## Usage of an Anonymization Solution at Courts in the Canton of Aargau, Switzerland

-------------

### Who is using it?
The High Court of the Canton of Aargau in Switzerland with overall roughly 470 employees of which 124 are using an anonymization solution.

### What problem(s) are they solving?
The court wanted to publish court decisions but needed to anonymize sensitive data before being able to do so. Therefore, they chose Balo.ai ([https://www.balo.ai](https://balo.ai/)) which allows for the anonymization of text documents while preserving the readability of the document and keeping all data on-premise. Through the usage of the anonymization solution, the court was able to publish a lot more decisions than before (1673 in one year as opposed to roughly 100 in the past).

### Which NLP technologies are they using?
Balo.ai works with an on-premise language model so that no data ever leaves the environment the tool is used in, which is particularly beneficial from a data protection perspective as the target users are mostly public institutions in the legal sector who operate under strict data protection regulations. The model is specifically trained for anonymization tasks and doesn't require a lot of computing resources, which makes it possible to host it on machines with limited resources. Additionally, there is a MS Word Add-In that makes anonymization suggestions that the user needs to manually accept and a separate tool for PDFs with the same functionality.

### Stage
Production - the courts are using the solution for anonymization

### Challenges
- The LLM is not able to understand all legal terminology and ideally needs to be fine-tuned with specific data for the customer
- Users would like a fully automated process which is not trivial
- Integration of AI-based tools into existing office software (e.g., through plugins) is not trivial as new versions require adaptations (also because many companies use no longer supported legacy software)

### Source
Two interviews conducted in April and May 2024, led by sebis researchers, one with a co-founder of Balo.ai and one with a test manager working for the courts of the Canton of Aargau.

Usage of an Anonymization Solution at Courts in the Canton of Aargau, Switzerland

## LLM Usage at a German Law Firm

-------------

### Who is using it?
The legal tech department of a German law firm is developing and using an AI-based solution for their daily legal work. The department is responsible for legal operations and tries to model legal processes as a series of steps that can be automated with software solutions.

### What problem(s) are they solving?
The motivation for the solution was an ever increasing influx of legal correspondence that took too much time sorting through and properly linking it to the respective cases.
Using an AI-based solution to go over all incoming correspondence automatically improves their efficiency since documents are classified and information like deadlines is extracted automatically.

### Which NLP technologies are they using?
The law firm uses Salesforce as CRM where they integrated their AI solution through multiple complex workflows. The solution Blaubach.ai ([https://www.blaubach.ai/](https://www.blaubach.ai/)) was specifically developed for internal use first but is now also offered to other interested parties and can be adapted to their specific needs.
The general workflow is that incoming data is passed through the AI solution that automatically detects the docket number and extracts relevant information from the document assigning it to the correct case.
The lawyers annotated the training data themselves and handed the labels off to the data scientists in the team to adapt the AI.

### Stage
Production - users are using the tool in their daily workflow

### Challenges
- The tool occassionally makes mistakes that need to be double-checked manually by humans
- Integrating new solutions into the existing legal IT landscape and adapting it to existing APIs is not trivial

### Source
Two interviews conducted in April 2024, led by sebis researchers, one interviewee is a lawyer and the head of the legal tech department at the company, the other is a data scientist.

LLM Usage at a German Law Firm

## LLM Usage at a German Law Firm

-------------

### Who is using it?
The legal tech department of a German law firm with over 350 employees is exploring the usage of AI-based solutions to support the daily legal work of the lawyers. The department consists of 6 employees and is responsible for the digital transformation and legal tech strategy.

### What problem(s) are they solving?
The most important problem to solve is how to schematically process large collections of documents. Therefore, extraction of relevant information from documents is the main use case the department is using AI-based solutions for. Other related use cases are the classification of documents and being able to ask the AI questions about the contents of a document.

### Which NLP technologies are they using?
The law firm has been using tools based on machine learning for multiple years and is currently looking at solutions like Legisway Analyzer ([https://www.wolterskluwer.com/en-gb/solutions/legisway/legisway-analyzer](https://www.wolterskluwer.com/en-gb/solutions/legisway/legisway-analyzer)) and Harvey ([https://www.harvey.ai/](https://www.harvey.ai/)). The output is perceived to be really good but still requires manual checking to make sure the information is not hallucinated.

### Stage
Production - users are using the tool in their daily workflow

### Challenges
- Most processes can not be fully automated yet and still require a human in the loop
- OCR does not always work as intended, especially for old documents that only exist on paper and need to be scanned
- Form-based data is more difficult to process with NLP than unstructured textual data (full sentences)
- Ensuring GDPR compliance

### Source
Two interviews conducted in May 2024, led by sebis researchers, one interviewee is a lawyer and the head of the legal tech department at the company, the other is an associate of the same law firm and part of the legal tech team.

## LLM Usage at a German Law Firm

-------------

### Who is using it?
A German law firm with roughly 180 employees has established a small legal tech lab and is exploring the usage of AI-based solutions to support their daily legal work. The department consists of 6 employees and is responsible for the digital transformation and legal tech strategy.

### What problem(s) are they solving?
The employees are trying to improve their knowledge management. The main goal is to make all relevant documents and data accessible through semantic search so that lawyers can easily find what they are looking for and then query these documents and receive appropriate responses from the AI. One example is to upload a document and ask for a summary of the decision. If the answer is not specific enough, the lawyer will ask a follow-up question to clear up any uncertainties. The users perceived big improvements in their efficiency as the solution is able to assist them in finding relevant information quicker than when looking for it themselves, especially considering that some cases consist of many documents which in turn consist of many pages. As LLMs may hallucinate responses, all users are urged to double check the generated responses.

### Which NLP technologies are they using?
The law firm is currently integrating Prime Legal AI ([https://primelegal.de/en/](https://primelegal.de/en/)) into their workflow. Depending on the use case, the AI solution utilizes multiple already existing LLMs (e.g., GPT-4, GPT-4-Turbo, Claude-3-Opus) under the hood to provide its various services. The users can select from a library of pre-defined prompts (or write their own) and the data they want them used upon. The anonymization functionality works on-premise whereas the rest of the functionality is available through the cloud. The service is accessible through a web application but also through an API. An entity-relationship model was used to train knowledge graphs which are the foundation for Prime Legal AI.

### Stage
Production - users are using the tool in their daily workflow

### Challenges
- Introduction of new tools requires training and marketing for the potential new users and ideally an IT department that takes care of the integration
- Data needs to be anonymized first before it can be fed into the tool
- Usage of AI leads to a paradigm shift from billing by the hour to a flat fee model

### Source
Two interviews conducted in May 2024, led by sebis researchers, one interviewee is the CEO of Prime Legal AI and the other is a lawyer specializing in corporate law and responsible for legal tech at their law firm.

## AI Usage at a German Law Firm

-------------

### Who is using it?

A large law firm with over 100 lawyers and accountants organized in teams of 10-12 employees distributed over 3 locations is in the process of evaluating legal AI usage at their firm. The target audience of the tools are the lawyers.
As of the date of the interview, the law firm has not yet adopted any company-wide legal AI solution but is testing various tools and models in a decentralized way to determine in which cases an adoption is feasible. Some lawyers already use personal ChatGPT accounts to assist them in their daily tasks.

### What problem(s) are they trying to solve?

The employees are in the phase of testing and figuring out what problems related to the legal domain can feasibly be solved with the usage of LLM-based tools.
Some lawyers already use personal ChatGPT accounts as a personal assistant to assist them in their daily tasks which are not restricted to the legal domain such as creating and summarizing texts (e.g., emails) or helping with translations, etc.
Regarding the legal use cases, due to the decentralized nature of the law firm the respective tasks of each department are diverse.
The use cases under investigation include but are not limited to:
- Document generation
- Document analysis of large(r) collections of documents
- Document summarization
- Information research and retrieval

### Which NLP technologies are they using?

The law firm is looking at various tools on the market, especially ones also being used by other German law firms, to get an overview of the current tool landscape. The solutions under investigation mainly operate with OpenAI and LLaMa models and are either deployed in the cloud or on-premise. In general, the law firm would prefer an on-premise solution to maintain control over the service. The language requirements are German and English. There is no intention of building their own tools but rather wait for the establishment of market standards that can then be adopted into their own IT landscape.

### How do they test the quality of the tools?

Since the rise of LLMs there have been multiple individual uncoordinated tests of LLM usage on a personal level but in the past few months an effort has been made to coordinate the testing and conduct it in a more organized way.
For this purpose, the various departments of the law firm individually test tools regarding their feasibility for their respective use cases. The employees do not use any specific evaluation metrics but come together and define success criteria and test scenarios for each task and then later on get together in working groups to discuss the findings.

### How is data privacy cared for?

No privacy-sensitive data is used for testing to avoid any privacy-related issues.

### Stage

Research - different users are experimenting with the usage of AI

### Challenges

- Commercial / Effort - For a revenue-driven firm the willingness to expend resources on adopting a new solution is limited as it requires additional effort for data preparation and testing before a solution can be integrated into the existing IT landscape.
- Privacy issues - Data fed into the model should not be used for training. For older data (historic data) there are no privacy agreements that allow the usage of the data in such tools.
- Compliance with AI Act

### Source
Interview conducted in February 2024, led by sebis researchers. The interviewees are working at the law firm in the position of lawyer and Digital Transformation Manager respectively.

AI Usage at a German Law Firm

## LLM Usage at a German Accounting Firm

-------------

### Who is using it?
The employees of a large German accounting firm with over 15.000 employees are exploring the usage of LLM-based tools to improve their efficiency.

### What problem(s) are they solving?
The employees use an LLM-based tool for document analysis, contract drafting, negotation assistance, and legal research. The most important use case for the users is document summarization, e.g., of contracts.

### Which NLP technologies are they using?
The employees are using Leah offered by ContractPodAI ([https://contractpodai.com/leah/](https://contractpodai.com/leah/)). ContractPodAI provides a contract lifecycle management suite and Leah is a standalone SaaS solution that enables the usage of generative AI without having to implement the full CLM solution. Leah provides various functionality out-of-the-box and utilizes multiple LLMs under the hood to generate its responses but it can also be fine-tuned to the needs of the customers.

### Stage
Production - users are using the tool in their daily workflow

### Challenges
- Continuous change management and data migration is required to keep up with new requirements
- Integration of the tool into the existing IT landscape
- Many lawyers are not very tech-savvy and require training to use the tool effectively

### Source
Interview conducted in May 2024, led by sebis researchers. The interviewee is the head of legal tech at the company.

LLM Usage at a German Accounting Firm

## LLM Usage at a German Law Firm

-------------

### Who is using it?
The corporate Merger & Acquistions department of a law firm totaling over 100 employees and specializing on German law is testing the usage of AI and LLMs to speed up their work.

### What problem(s) are they solving?
The employees use an LLM-based tool in hopes of improving their efficiency to draft contracts. Normally, new contracts are constructed by reusing and adapting clauses from past contracts. The tool helps automate this task by importing clauses from other documents and adapting them to the new context. The tool is used for assistance as it still requires a good amount of manual effort.
One reason for incorporating AI technology into their weekly routine is the perception that customers now expect it in this day and age.

### Which NLP technologies are they using?
A few selected lawyers of the department are testing Henchman ([https://henchman.io/](https://henchman.io/)), a multi-LLM contract drafting and negotiating AI solution, to assess a potential department-wide adoption of the tool. One of the employed LLMs is ChatGPT running on an Azure instance so that the data is not used to train the model and remains within the EU. Users can interact with the system through an Add-In in Word or Outlook.

### Stage
Research - different users are experimenting with the usage of LLMs

### Challenges
- Lawyers need to invest time and effort to incorporate the tool into their routine before any efficiency gains can be realized
- The tool is only as good as the existing documents from the past. Changes in legal regulations are naturally not reflected yet for future contracts.

### Source
Two interviews conducted in April and May 2024, led by sebis researchers. One interviewee is a lawyer working in corporate Mergers and Acquisitions at a German law firm and the other is a legal tech consultant.

## LLM Usage at a German Technology and Engineering Company

-------------

### Who is using it?
The patent department of a global technology and engineering company, which includes 26 direct employees and around 200 in total, is testing the usage of AI and LLMs to speed up their work.

### What problem(s) are they solving?
The employees use the LLM for various tasks surrounding text generation and analysis. Since September 2023, they are testing what is possible concerning patents including patent drafting, mapping of patents, and patent analysis, i.e., retrieving key aspects out of the documents. The results are perceived to be quite promising but one still needs to invest some time refining them as they are not good enough to just be copy pasted and used as is.

The initial expectation was that you can just give the LLM a job and wait for it to come back with the results. However, one key learning is that you need to teach the model what you want, i.e., give it a persona and break down the instructions into very small parts.

Further use cases revolve around assistance in their daily jobs such as drafting emails or parts of reports and translations.

In the future, they would like to expand their usage of LLMs by leveraging the voice input capabilities. Ideally, the LLM will act as a personal assistant helping with their daily tasks, e.g., answering questions, providing explanations, finding and opening files, etc.

### Which NLP technologies are they using?
The employees have access to ChatGPT 3.5 with 128k tokens through the company's Azure instance which provides a certain level of security and confidentiality.
The department is considering the costs and benefits of prompt optimization or the potential fine-tuning or even training of own models.

### Stage
Research - different users are experimenting with the usage of LLMs

### Challenges
- Context window insufficient - 128k tokens are not enough for working with large documents
- Domain-specific knowledge required - LLMs are not specialized enough for such a technical domain especially when it comes to novel technology that is not part of training data
- Fine-tuning / Training models requires necessary know-how and good training data

### Source
Interview conducted in June 2024, led by sebis researchers. The interviewee is a lawyer and head of the patent department at the abovementioned company.

LLM Usage at a German Technology and Engineering Company

## LLM Usage at a German Healthcare Solution Provider

-------------

### Who is using it?

A team of several in-house legal experts from a multi-national, German-based technical corporation specializing in healthcare solutions and services. 300 users are currently testing the usage of MS Copilot.

### What problem(s) are they solving?

The employees use LLM-based tools for various tasks in their daily work life. For one as a personal assistant for various tasks including quick general questions, translations, summaries of text, and searching for specific files on the system. Another task can be summed up under the umbrella term of content generation, e.g. for email drafts, marketing texts, etc.

The rise in popularity of ChatGPT led legal experts at the company to evaluate if they could also harness the power of LLMs to assist in the daily, routine work of legal contract generation and analysis, which typically requires a great deal of manual time and human effort. The usage of LLMs can help highlight important parts of contracts and give guidance on where a human expert should have a closer look. The contracts are based primarily on German law. Other legal tasks are also being considered.

Another emerging use case is the usage of LLMs to assist in software development wherefore it's currently being tested regarding legal regulations.

Finally, the organization is thinking about deploying chatbots to assist in the communication with customers which could result in a reduction of costs.

### Which NLP technologies are they using?

As a technical-oriented organization, employees already had existing access to the Microsoft Office products, which facilitated for the convenient experimentation with MS Copilot and ChatGPT on the Microsoft Azure Cloud. This cloud solution was chosen due to its ease of use, as well as the perceived privacy of using ChatGPT in the cloud, with no need for further training or fine-tuning on proprietary data. Prompt engineering is employed to optimize results.

Software engineers are also evaluating open-source models for generating code and documenting it automatically.

### How do they test the quality of their service?
The organization started using the ChatGPT playground in summer 2022. Before that, interested parties experimented with their personal ChatGPT accounts and non-sensitive data, subject to guidelines provided by the organization. The employees do not employ any prompt engineering guides on department level or use any evaluation metrics to determine the effectiveness of the model. The experts manually test the models for their various tasks to determine if their usage for the respective use cases makes sense. One finding is that the more the respective field is legally regulated, the less sense it makes to employ AI, e.g. German lease contracts are too strictly regulated so that LLMs struggle with producing accurate and correct results.

### How is data privacy cared for?
Documents are anonymized manually before uploading them to the cloud and making them accessible to the model.
Usage of the public ChatGPT interface is only allowed for non-privacy sensitive matters. For everything else the ChatGPT playground needs to be used as Microsoft claims to not use that data for training purposes.

### Stage
Research - different users are experimenting with the usage of LLMs

### Barriers to Adoption
- Hallucinations -– Quality, reliability, and correctness of generated responses varies depending on use case.
- Copyright infringement and ensuing liability risk - It's not always clear whether generated responses from the model contain copyrighted content.
- Privacy issues - Data fed into the model should not be used for training.


### Source
Interview conducted in January 2024, led by Prof. Florian Matthes and Nektarios Machner. The interviewee is a lawyer and responsible for the IT department at the abovementioned company, where he also looks into what is going on in the IT world regarding useful tools.


LLM Usage at a German Healthcare Solution Provider

## AI Solution Provider: Elephant Labs

-------------

### Who is using it?
The solution is targeted at public institutions like the government, courts, and other regulatory offices. The company is currently in contact with the justice ministry of Bavaria to explore how the solution can provide value to them.

### What problem(s) are they solving?
The company Elephant Labs ([https://www.elephantlabs.ai/](https://www.elephantlabs.ai/)) was founded in December 2022 in an attempt to develop a solution that increases the efficiency of lawyers through the use of computer science. For example, court verdicts need to be published to the public but the regulations as to how and when are not clear. In order to publish such data, however, sensitive information needs to be masked first and this is what VerdictLLM was initially developed for.

### Which NLP technologies are they using?
The tool VerdictLLM utilizes the power of open-source LLMs and statistical methods to anonymize sensitive data. Additionally, the company uses embeddings to classify which information needs to be anonymized. The LLM is fine-tuned through synthetic data specifically optimized for applicability in the legal context.

### Challenges
- Establishing trust in LLM-based solutions with potential customers who primarily work with sensitive data
- Public institutions are reluctant to change existing processes (independent of AI)

### Source
Interview conducted in April 2024, led by sebis researchers. The interviewee is the founder of the company.

AI Solution Provider: Elephant Labs

## AI Solution Provider: fastlaw

-------------

### Who is using it?
The solution is targeted at law firms and notaries working with land registration data.

### What problem(s) are they solving?
The company fastlaw ([https://fastlaw.online/](https://fastlaw.online/)) was founded as a spin-off from a law firm that has been working completely digitally for the past ten years alreay. Their AI solution is designed to read land register data and extract relevant information.

### Which NLP technologies are they using?
The company provides a CRM solution for law firms and an AI-based tool called GREGOR available as a SaaS solution that is able to read land register data through OCR and then extract relevant information which in turn can be exported again in a structured format. To achieve this, the company together with an undisclosed partner trained their own language model.

### Challenges
- Adapting the AI to correctly identify the relevant text passages
- Land register data is not standardized and can consist of multiple document excerpts assembled into one single document

### Source
Interview conducted in April 2024, led by sebis researchers. The interviewee is the head of legal engineering and business development of the company.

AI Solution Provider: fastlaw

## LLM Usage at a German Law Firm

-------------

### Who is using it?
A large German law firm is developing their own solution to assist in their legal work but the solution is also provided to interested legal professionals who need to generate documents in their daily work, e.g., contracts.

### What problem(s) are they solving?
The main use case aims at generating legal documents based on templates and data inputs but the solution is also able to assist in other legal tasks surrounding document management. The company is currently evaluating the usage of AI in various aspects of their work to derive new use cases that can benefit from its usage.

### Which NLP technologies are they using?
The company uses AI pipelines (i.e., Retrieval Augmented Generation) that are independent of a specific LLM and always aim to use the best LLM depending on use case. Currently, they are working with GPT-4 deployed on their own Azure instance but are aiming at also including Claude Opus. The company took all their documents, anonymized them, and embedded them in a vector database to allow for a semantic search. From this existing knowledge base new documents can be generated through pre-defined prompts. The functionality is accessible through a Word Add-In that is connected to the pipeline built on the deepset cloud. Subtasks of the pipeline also employ other NLP methods such as Named Entity Recognition (NER).


### Stage
Research - the solution is under active development and different users are evaluating its usage

### Challenges
- Developing your own AI is resource-intensive and not practical in an environment with frequent law changes
- Users require proper training in order to utilize the full potential of novel AI-based solutions

### Source
Interview conducted in April 2024, led by sebis researchers. The interviewee is the CEO of a spin-off company.

## AI Solution Provider: Herlock Insights

-------------

### Who is using it?
The solution is targeted at law firms and legal professionals.

### What problem(s) are they solving?
The main focus is to provide transparency regarding legal case data, i.e., every legal case is self-contained and builds the foundation that the software solution operates upon. Users can query the documents for information and ask about information related to each case.

### Which NLP technologies are they using?
The company provides an AI-based case analysis tool called Herlock.ai ([https://www.herlock.ai/](https://www.herlock.ai/)) that uses natural language processing (NLP) and vector embeddings to analyze legal case data. The solution is provided as SaaS deployed on a private cloud running on GCP or AWS. One of the used language models is ChatGPT but the company is looking at alternatives that can be deployed on-premise for customers that have this requirement. The solution organizes relevant information around individual legal cases and provides a visual timeline of chronologically sorted events next to allowing to search and compare documents and ask questions about them.

### Challenges
- Hallucinations can occur but are mitigated by providing references to the sources of the information
- Legal landspace is very diverse and many users are still uncertain about employing AI in their workflow

### Source
Interview conducted in April 2024, led by sebis researchers. The interviewee is the co-founder and CTO of the company.

AI Solution Provider: Herlock Insights

## AI Solution Provider: Hyde

-------------

### Who is using it?
The solution is targeted at law firms and legal professionals.

### What problem(s) are they solving?
The motivation to focus on the legal sector stems from the fact that legal professionals primarily work with text which is predestined to being used with large language models. The offered AI suite aims to assist with multiple legal use cases including case management, document analysis, and document generation based on custom templates specific to the customer.

### Which NLP technologies are they using?
Hyde ([https://www.hyde.to/](https://www.hyde.to/)) is an AI suite based on language models. The company transitioned from building one large language model into building multiple smaller models specific for individual use cases such as semantic search to query documents for information. The solution is offered as SaaS in the cloud or on-premise depending on the customer's need.

### Challenges
- Customer acquisition is not easy in the legal domain as law firms are focused on their work and tend to not have much time for sales calls

### Source
Interview conducted in May 2024, led by sebis researchers. The interviewee is a late co-founder and the CPO at the company.

AI Solution Provider: Hyde

## AI Solution Provider: Justin Legal

-------------

### Who is using it?
The solution is targeted at law firms and legal professionals.

### What problem(s) are they solving?
The company focuses on data collection and data quality to enable digitalization. Clients of law firms can use the solution to provide relevant data for their case in a structured form by following the steps in the tool. Lawyers are then able to review the data digitally and ask questions about the document to the AI assistant.

### Which NLP technologies are they using?
Justin Legal ([https://justin-legal.com/](https://justin-legal.com/)) provides their service through a SaaS web application. The underlying LLM is ChatGPT 4 deployed on an Azure instance which is used to provide features like text summarization and question & answering. The data can be used to generate new documents, export in various formats, or directly passed via API to standard software solutions.

### Challenges
- Properly modeling legal real-world scenarios in a digital system
- GDPR Compliance

### Source
Interview conducted in May 2024, led by sebis researchers. The interviewee is the CEO at the company.

AI Solution Provider: Justin Legal

## AI Solution Provider: Libra Tech

-------------

### Who is using it?
The solution is targeted at law firms and legal professionals.

### What problem(s) are they solving?
Libra ([https://libratech.ai/](https://libratech.ai/)) wants to become an AI-powered law firm assistance provider to increase the efficiency of lawyers. For example, the tool is able to classify and extract information from documents and generate new documents like contracts.

### Which NLP technologies are they using?
The company is currently building modules in an AI pipeline to automate different tasks in a lawyer's workflow. They started out with using commercial APIs like ChatGPT but then proceeded to fine-tune their own models that will be used exclusively in the future. The solution is under active development and the company is building their own evaluation data sets while simultaneously receiving feedback from customers to evaluate their application.

### Challenges
- Compliance requirements
- Acquiring high-quality training data to fine-tune models to the legal domain
- Aligning the technical side with the business side

### Source
Interview conducted in April 2024, led by sebis researchers. The interviewee is a co-founder and the CTO of the company.

AI Solution Provider: Libra Tech

## AI Solution Provider: Mioto Labs

-------------

### Who is using it?
The solution is targeted at small- to medium-sized law firms.

### What problem(s) are they solving?
Mioto Labs ([https://mioto.app/](https://mioto.app/)) offers custom-tailored solutions for customers in the legal domain. Typical use cases include but are not limited to contract generation based on pre-defined templates, document review and analysis, and information extraction.

### Which NLP technologies are they using?
The solution currently consists of a workflow builder for process automation. Commercial LLMs like ChatGPT through an Azure instance are used to assist in specific steps of the workflow, e.g., to extract information from documents. Depending on the needs of the customers, the team decided where and when in the workflow chain it makes sense to utilize LLMs to achieve a specific subtask.

### Challenges
- SaaS is difficult to market to customers in the legal domain
- Building custom solutions for customers requires a lot of flexibility from the development team as requirements are not often clear at the beginning

### Source
Interview conducted in May 2024, led by sebis researchers. The interviewee is a co-founder and the CTO of the company.

AI Solution Provider: Mioto Labs

## AI Solution Provider: Neur.on AI

-------------

### Who is using it?
The solution is targeted at professionals from the legal and financial sector.

### What problem(s) are they solving?
The company Neur.on AI ([https://neur-on.ai/](https://neur-on.ai/)) specializes in translations and translation management.

### Which NLP technologies are they using?
The company currently provides multiple separate solutions like Lex Machina and Corrext that they eventually want to integrate into one overarching software solution. The solution is offered as a collaborate SaaS solution accessible through a web application. The system preprocesses data with the help of multiple LLMs, e.g., to categorize what kind of law documents pertain to, and then the system can provide machine translations but also humans can collaborately check the results and work on creating the translations through a web interface manually. The models were trained specifically for legal and financial terminology.

### Challenges
- Ensuring confidentiality of the data

### Source
Interview conducted in May 2024, led by sebis researchers. The interviewee is the COO at the company.

AI Solution Provider: Neur.on AI

## AI Solution Provider: 913.ai

-------------

### Who is using it?
The solution is targeted at legal service providers.

### What problem(s) are they solving?
913.ai ([https://www.913.ai/](https://www.913.ai/)) wants to provide an automation infrastructure that allows users (e.g., legal engineers) to connect various already existing tools in a law firm with each other. Use cases include but are not limited to document classification and information extraction. The solution aims to represent and assist in typical workflows in the daily legal work of lawyers.

### Which NLP technologies are they using?
913.ai uses open-source LLMs but also fine-tunes their own models for specific tasks like text extraction or classification. They offer a pipeline builder that allows users to select and link various modules together to depict their own unique scenarios as they need them. They are currently adding more modules and improving already existing ones to make them more adaptable by the users.

### Challenges
- GDPR Compliance
- Gaining new customers as customers in the legal domain tend to be more averse to cloud solutions

### Source
Interview conducted in April 2024, led by sebis researchers. The interviewee is a lawyer, co-founder and the CEO of the company.

AI Solution Provider: 913.ai

## AI Usage at a German Law Firm

-------------

### Who is using it?

A large law firm with over 300 lawyers is evaluating the use of legal AI at their firm. The target audience of the tools are legal professionals.

### What problem(s) are they trying to solve?
The firm is interested in understanding the potential benefits of using AI in their practice, such as automating repetitive tasks, improving decision-making speed, and increasing overall efficiency. Therefore, the firm called into action an AI initiative consisting of multiple lawyers to explore and evaluate the use of legal AI.

The use cases under investigation include:
- Document analysis
- Document generation (e.g., contracts)
- Document summarization (e.g., summarizing court decisions)
- Due diligence
- Question Answering

For Question Answering, they are building a curated knowledge base containing all admissible company data that can then be queried through the chat interface. The semantic search also provides references to the sourced documents.

### Which NLP technologies are they using?
The law firm is evaluating the use of different LLMs including proprietary models like ChatGPT (OpenAI) and Luminous (Aleph Alpha), as well as open-source models like Llama.
For their daily work, all employees have access to ChatGPT deployed on their own Azure instance to ensure data is not used for training. To manage and reduce the costs associated with LLM usage, token length is limited.
Some users are also testing Copilot for their daily work needs.

### Stage
Production - different users are utilizing AI technology regularly in their workflows

### Challenges
- Costs of AI usage are not insignificant and need to be managed
- Employees need to be properly trained to ensure compliant usage of the AI solutions and maximize their usefulness
- The speed at which AI technology advances makes it difficult to keep up, e.g., context window increases (tokens)
- Societal implications, e.g., risk of fraud and scams

### Source
Interview conducted in March 2024, led by sebis researchers. The two interviewees are both lawyers at the law firm.

## LLM Usage at a Global Law Firm

-------------

### Who is using it?
The Mergers and Acquistions department of a large international law firm with roughly 70 lawyers in Germany (900 globally) is exploring the usage of AI in their daily work. The law firm has a global technology committee consisting of multiple employees including the CIO that evaluates new technologies and makes executive decisions about adopting them into the firm.

### What problem(s) are they solving?
The employees are using AI-based tools for document analysis, information extraction, and translation. They believe AI tools will eventually replace a lot of the work that is nowadays being done by junior associates.

### Which NLP technologies are they using?
The department is using Kira ([https://kirasystems.com/](https://kirasystems.com/)) for document analysis and information extraction from legal documents like contracts. The firm has access to its own instance running the software to ensure the confidentiality of the data.
For translations the lawyers use DeepL ([https://www.deepl.com/](https://www.deepl.com/)), an AI-based translation tool specifically trained for translation tasks.
Furthermore, the firm also tested ChatGPT but deemed it not good enough to be used as-is in their daily professional work and is now evaluating Harvey ([https://www.harvey.ai/](https://www.harvey.ai/)) for document analysis.

### Stage
Production - AI-based tools are used regularly in their legal work

### Challenges
- Ensuring GDPR compliance and confidentiality
- Trust in AI systems needs to be established
- Support for multiple languages is important for an international firm
- Not enough use cases (at the time of the interview) to justify large investments in AI
- Rethinking of business model (hourly billing vs. flat fee) might be necessary if AI increases efficiency

### Source
Interview conducted in February 2024, led by sebis researchers. The interviewee is a lawyer working in Mergers and Acquistions at a large law firm.

LLM Usage at a Global Law Firm

## AI Usage at a German Publishing Group

-------------

### Who is using it?
A large publishing group with its own legal department consisting of 6 lawyers is evaluating the usage of AI tools for their daily legal work.

### What problem(s) are they trying to solve?
The employees have access to a company-wide chatbot that was developed in-house and that they can use for their daily work. The use cases include:
- Document analysis
- Document summarization
- Text generation (e.g., email, rephrasing)
- Translation
- Question Answering
- Research

### Which NLP technologies are they using?
The company chatbot is based on models from OpenAI but does not have access to the internet or company data directly. However, users can upload documents through the chatbot interface to use them in ensuing prompts.
The company is also testing Semantha ([https://www.semantha.de/](https://www.semantha.de/)) for document analysis and semantic search.

### How do they test the quality of the tools?
The legal professionals do not have a predefined set of evaluation metrics but come together regularly to discuss the usage of AI and other relevant topics revolving around AI. The general sentiment is that there is a lot of potential in AI solutions but we are still just at the beginning and the tools currently do not work satisfactorily enough yet.

### Stage
Production - employees use company-wide chatbot for their daily work

### Challenges
- Hallucinations - LLMs do not always produce accurate information
- Finding the right AI solution(s) for each use case
- Ensuring that employees understand how to use the chatbot effectively through proper training
- Ensuring that company data is not used for AI training

### Source
Interview conducted in March 2024, led by Prof. Dr. Florian Matthes and Nektarios Machner. The interviewee is working at the firm's legal department as a lawyer with a focus on IT.

AI Usage at a German Publishing Group

## AI Usage at a German Law Firm
-------------

### Who is using it?
A German law firm with roughly 120 lawyers is exploring the use of AI in their daily legal work. The firm put together an AI Task Force consisting of multiple lawyers from the firm to evaluate AI-based tools regarding their applicability in their daily workflow.

### What problem(s) are they trying to solve?
The main use cases are document analysis and document generation, e.g., for documents submitted by a party in a legal proceeding or to create pitch decks. Further use cases include translation and general research. To assist with cases concerning intellectual property, an AI-based tool is used to do similarity analysis, e.g., comparing two logos or design.

Overview of use cases:
- Document analysis
- Document generation
- Translations
- Research
- Similarity Analysis

### Which NLP technologies are they using?
The law firm is currently evaluating ChatGPT to find out what for and how it can be used in their legal work. Additionally, Microsoft's Copilot is being explored as a personal assistant, e.g., to search for specific files on the file system. DeepL Pro ([https://www.deepl.com/](https://www.deepl.com/)), an AI-based translation tool specifically trained for translation tasks, is used for translations.

### How is data privacy cared for?
No privacy-sensitive data is used for testing to avoid any privacy-related issues.

### Stage
Production - lawyers use AI-based tools for their daily legal work

### Challenges
- Hallucinations - LLM output cannot be used as-is but needs to be double-checked and verified
- A lot of interesting AI solutions are put on the market but there is no proper overview to keep track of developments
- GDPR compliance, privacy and confidentiality need to be ensured
- AI solutions need to be integrated properly into the existing IT landscape and processes, which is not trivial for law firms without or with only a small IT department

### Source
Interview conducted in February 2024, led by sebis researchers. The interviewees are working as lawyers at the law firm.

# Legal AI Usage at German Law Firm

-------------

### Who is using it? ###
M&A lawyers and other legal professionals at a large German law firm are using legal AI tools, particularly within the context of document-heavy legal work such as due diligence and contract review. Legal tech professionals and lawyers with technical backgrounds support development and adoption.

### What problem(s) are they trying to solve? ###
They aim to reduce the time and effort required for document classification, summarization, and legal analysis during tasks like due diligence and litigation. AI is used to extract key contractual clauses, generate executive summaries, and assist with internal legal process improvements. There's also a focus on improving access to templates and internal resources via AI-assisted interfaces.

### Which NLP technologies are they using? ###
They are using large language models (LLMs), particularly GPT-4 and GPT-4o, in both internal (on-premise or country-hosted) and third-party tools. Cloud-based models are used cautiously and only when permitted. Tools include proprietary LLM-based chat interfaces, as well as domain-specific models like those integrated with legal publishers' databases. They also experiment with prompt engineering and use explanation-focused AI tools for model transparency.

### Stage ###
Production

### Challenges ###
- Hallucination risks in LLM outputs requiring careful review
- Low adoption rates due to lack of user understanding and training
- High effort for structuring internal data for AI usage
- Difficulties in aligning legal and technical teams
- Market pressure to reduce billable hours via AI without compromising quality
- Limitations in accessing comprehensive case law databases
- Lack of transparency in AI decision-making compared to human associates
- Change management issues and resistance from staff

### Source ###
Interview conducted in February 2025, led by sebis researchers. The interviewee is a legal professional working at the organization.

Legal AI Usage at German Law Firm

# Legal AI Usage at German Technology Company

-------------

### Who is using it? ###
The legal department of a German technology company, consisting of eight in-house lawyers, is exploring the use of legal AI tools. Their work includes corporate law, M&A, and international contracts.

### What problem(s) are they trying to solve? ###
They aim to improve efficiency in handling legal documents and contracts. Specific use cases include translating legal texts, rephrasing clauses, summarizing lengthy documents (e.g., EU regulations), and creating initial drafts of legal clauses. A key goal is to automate the review and negotiation of standardized non-disclosure agreements (NDAs) to reduce lawyer involvement in routine processes.

### Which NLP technologies are they using? ###
They currently use on-premise chatbots based on open-source models such as LLaMa and Mixtral. Cloud-based models like ChatGPT are not permitted due to company policy. The department also experiments with proprietary chatbots embedded in legal databases (e.g., beck-online). Tools like DeepL are used for translation, and internally developed contract management systems support their workflow.

### Stage ###
Research

### Challenges ###
- Cloud-based AI models are blocked due to strict internal IT policies.
- Existing commercial tools were too costly relative to their benefits.
- Most contracts are unique and complex, limiting automation potential.
- High effort required to create playbooks or training data for AI tools.
- On-premise models lack legal domain training and access to legal databases.
- Technical limitations with document length and functionality.
- Need for better prompt engineering knowledge; planning internal training.
- Current tools cannot cite legal sources reliably.
- Manual verification of AI outputs is still necessary due to accuracy concerns.

### Source ###
Interview conducted in February 2025, led by sebis researchers. The interviewee is a legal department manager and a legal counsel, working at the organization.


Legal AI Usage at German Technology Company

# Legal AI Education for German Legal Professionals

-------------

### Who is using it? ###
The users are primarily legal professionals in Germany, including lawyers, legal departments, and law students. Small law firms and individual practitioners show the most active interest.

### What problem(s) are they trying to solve? ###
Users aim to understand and adopt AI tools to improve efficiency in daily legal work. Specific problems include contract analysis, drafting and summarizing legal emails, document review, legal research, and time tracking for billing. The goal is to develop AI literacy and practical skills for integrating AI into legal workflows.

### Which NLP technologies are they using? ###
They use general-purpose models like ChatGPT and Perplexity, often in secure or self-hosted environments due to privacy concerns. Tools that mimic ChatGPT functionality but claim compliance with professional and data protection standards are also in use. Retrieval-Augmented Generation (RAG) systems are noted for document analysis tasks.

### Stage ###
Research

### Challenges ###
- Legal sector's general resistance to change and risk aversion  
- High skepticism regarding data protection and regulatory compliance  
- Lack of structured AI knowledge and technical expertise among lawyers  
- Difficulty in motivating adoption and routine use of AI tools  
- Structural inefficiencies and traditional workflows, especially in courts  
- Cost barriers for smaller firms  
- Media discontinuity and inconsistent software standards across jurisdictions

### Source ###
Interview conducted in February 2025, led by sebis researchers. The interviewee is a legal tech entrepreneur and former product manager, working at the organization.

Legal AI Education for German Legal Professionals

# Legal AI Usage at German Consulting Firm

-------------

### Who is using it? ###
Legal AI tools are used by consultants and internal teams within the firm to support both client projects and internal legal workflows. The tools are also introduced and tested in client organizations through advisory projects.

### What problem(s) are they trying to solve? ###
The primary goals are to support text-based legal tasks such as document drafting, analysis, summarization, and information extraction. The firm also helps clients identify relevant use cases and integrate AI tools into their legal processes through workshops and proofs of concept. Legal AI is not seen as replacing legal professionals but as a tool to assist them.

### Which NLP technologies are they using? ###
The organization uses general-purpose generative AI models, particularly ones based on ChatGPT, with additional internal layers for customization. Tools mentioned include internally hosted generative AI chat systems and specialized legal AI assistants. Some tools rely on fine-tuning or prompt engineering to better address legal contexts. Most tools are cloud-based, though internal deployments are preferred for data-sensitive tasks.

### Stage ###
Production

### Challenges ###
- High user expectations vs. actual capabilities
- Identifying relevant use cases for each user group
- Integration into existing workflows
- Misunderstandings about what AI can and cannot do
- Issues related to hallucinations and knowledge scope
- Requirement for user education and change management
- Configuration complexity for specialized tools
- Slow adoption in conservative markets like Germany

### Source ###
Interview conducted in January 2025, led by sebis researchers. The interviewee is a legal technology consultant, working at the organization. 

Legal AI Usage at German Consulting Firm

# Legal AI Usage at German Law Firm

-------------

### Who is using it? ###
Lawyers, assistants, and secretarial staff across multiple practice groups at a German law firm are using legal AI tools. The digital strategy is coordinated centrally, with adoption managed and tracked across departments.

### What problem(s) are they trying to solve? ###
The firm aims to improve efficiency in legal workflows, particularly:
- Preparing case summaries for litigation
- Drafting documents and legal texts
- Proofreading and summarizing materials
- Internal knowledge management and document indexing

A core focus is on reducing time spent on manual processes and supporting legal work with AI-assisted summaries and drafting components.

### Which NLP technologies are they using? ###
The firm uses a multi-tiered approach:
- General-purpose chatbots (e.g., ChatGPT, Copilot) for experimentation and basic tasks like proofreading and drafting.
- More secure, legal-specific AI tools based on LLMs (some derived from ChatGPT) for sensitive tasks.
- Cloud-based solutions with a fallback to EU-hosted tools for data-sensitive clients.
- Ongoing pilot testing with additional models, including those for knowledge management.

They do not use enterprise licenses for public LLMs like ChatGPT but instead enforce strict usage guidelines and access restrictions.

### Stage ###
Production

### Challenges ###
- Hallucinations and inaccuracies in LLM outputs
- Poor performance of certain tools (e.g., Copilot in EU version)
- Low adoption or hesitation among conservative users
- Need for strict compliance with data protection laws (mandating internal usage guidelines and access gates)
- Difficulty achieving full-text draft quality with LLMs; users rely on building block-style drafting
- Scalability of custom tool development is limited due to firm size
- Managing and motivating sustained tool adoption across departments

### Source ###
Interview conducted in February 2025, led by sebis researchers. The interviewee is a digital transformation lead, working at the organization.

# Legal AI Usage in Swiss Document Anonymization Software

-------------

### Who is using it? ###
The users include court departments, legal authorities, and administrative bodies ranging from small teams of five people to larger organizations with up to 200 employees.

### What problem(s) are they trying to solve? ###
The primary use case is the anonymization and pseudonymization of legal documents, such as court decisions. The goal is to automate suggestions for redactions to reduce manual effort, especially in large documents. Clients also seek to handle special formatting challenges in PDFs and accommodate jurisdiction-specific identifier structures, like those found in Austrian property designations.

### Which NLP technologies are they using? ###
They use on-premise NLP models for named entity recognition. The system is not based on generative models and does not rely on cloud services. The technology stack includes a base model trained on pseudonymized court decision data and additional logic using pattern matching (e.g., regex). They also make use of open-source components like spaCy.

### Stage ###
Production

### Challenges ###
- Complex formatting and non-machine-readable content in PDFs
- Balancing precision and over-redaction based on diverse client needs
- Limitations of OCR quality in source documents
- Client IT constraints and long deployment timelines
- High expectations due to public familiarity with tools like ChatGPT
- Legal-specific data handling requirements and slow cloud adoption in public institutions
- Resource limitations for small clients and aversion to subscription models

### Source ###
Interview conducted in June 2025, led by sebis researchers. The interviewee is a managing director, working at the organization.

Legal AI Usage in Swiss Document Anonymization Software

# Legal AI Usage by a Self-Employed Legal Professional in Germany

-------------

### Who is using it? ###
A self-employed legal professional with a background as both a lawyer and Legal Engineer, currently running her own law firm and advising other firms on digitalization.

### What problem(s) are they trying to solve? ###
AI is being used to streamline marketing and content creation (e.g., blog posts and videos), assist with legal research and brainstorming, and draft client communications. Additionally, the interviewee advises other firms, such as tax or compliance-focused organizations, on how to implement legal tech solutions to optimize workflows and automate repetitive tasks.

### Which NLP technologies are they using? ###
The interviewee uses ChatGPT (GPT-based) for brainstorming, writing support, and generating explanatory content. Legal research tools include AI-powered databases that summarize court decisions. There is experimentation with cloud-based tools, some of which also allow user-uploaded documents. Technologies mentioned are generally cloud-based.

### Stage ###
Research

### Challenges ###
- GPT models produce unreliable legal arguments and incorrect formatting in legal documents.
- High costs of legal research platforms limit accessibility for solo practitioners.
- Legal documents often cannot be shared with AI tools due to data protection concerns.
- Difficulty anonymizing court decisions for upload or use in tools.
- General lack of digital and AI readiness in parts of the legal profession.
- Resistance to change within firms due to psychological and organizational inertia.
- Many legal AI tools lack context awareness or require human oversight.

### Source ###
Interview conducted in February 2025, led by sebis researchers. The interviewee is a self-employed legal professional with consulting experience in legal tech.

Legal AI Usage by a Self-Employed Legal Professional in Germany

# Legal AI Usage at German Law Firm

-------------

### Who is using it? ###
Lawyers at a large international law firm, especially those in areas like M&A, Private Equity, Real Estate, and Finance, are using legal AI tools. Usage varies by practice area and seniority.

### What problem(s) are they trying to solve? ###
They aim to improve efficiency in tasks such as document review, translation, and drafting standard contracts. AI is used for due diligence, generating draft memos, and benchmarking market standards. While high-end, bespoke legal work remains largely manual, AI supports standardization and consistency in more repetitive or commodity-oriented tasks.

### Which NLP technologies are they using? ###
They use general-purpose models such as ChatGPT and GPT-4, as well as a mix of external and proprietary tools. Some tools are integrated into cloud-based systems; others are custom-built or customized in-house. AI is employed for language translation, summarization, and drafting assistance, though bespoke legal advice is not generated by AI.

### Stage ###
Production

### Challenges ###
- Reluctance to share proprietary legal knowledge, even anonymized
- AI's limitations in handling complex, bespoke contracts
- Difficulty aligning AI-generated output with client-specific expectations
- High cost and maintenance of in-house tool development
- Concerns about consistency, accuracy, and overreliance on averages or correlations
- Resistance from experienced lawyers in high-end legal areas
- Difficulty in monitoring and evaluating numerous third-party legal tech vendors
- Legal industry’s emphasis on short-term profitability hampers long-term AI investments

### Source ###
Interview conducted in January 2025, led by sebis researchers. The interviewee is a senior lawyer, working at the organization.

# Legal AI Usage at German Law Firm

-------------

### Who is using it? ###
Lawyers at a large international law firm in Germany, particularly those in practice groups and legal tech roles, are using generative AI tools. A dedicated Legal Tech Fellow supports the integration and training of these tools across the firm.

### What problem(s) are they trying to solve? ###
They use legal AI tools for drafting documents (e.g., emails, memos, court pleadings), translating texts, explaining technical terminology, conducting preliminary legal research, and summarizing lengthy legal documents. The goal is to improve efficiency, reduce manual workload, and make legal knowledge more accessible across departments.

### Which NLP technologies are they using? ###
They primarily use Microsoft Copilot integrated with Office 365. Other tools such as ChatGPT and Perplexity have been tested, though Perplexity is used only privately due to data restrictions. They also co-develop specialized tools in collaboration with tech providers, such as a model for identifying legal risks in internal documents. Most tools are cloud-based, with strict attention to data protection and legal compliance.

### Stage ###
Production

### Challenges ###
- Difficulty integrating AI tools with existing systems
- Concerns over data security, especially with client data
- Limited accuracy in legal content generation (e.g., hallucinated citations)
- Prompting complexity and low intuitiveness for new users
- High effort required for data anonymization in model training
- Legal and regulatory constraints, especially in internal investigations
- Limited document upload sizes and processing constraints

### Source ###
Interview conducted in February 2025, led by sebis researchers. The interviewee is a Legal Tech Fellow and practicing lawyer, working at the organization.


# Legal AI Usage at German Law Firm

-------------

### Who is using it? ###
Lawyers across different departments, especially in tax compliance, employment law, and real estate transaction teams. Some legal assistants and administrative staff also use AI-assisted tools for translation and reporting tasks.

### What problem(s) are they trying to solve? ###
The primary goals are to increase efficiency and reduce manual labor in document analysis, contract drafting, legal research, and translations. AI tools are used to assist with extracting data from large volumes of legal and financial documents, generating and managing standard contract clauses, improving translation quality, and drafting standard communications.

### Which NLP technologies are they using? ###
They are testing or using tools based on large language models like ChatGPT, especially through integrations with publishers (e.g., Otto Schmidt, Beck). Some tools operate via browser-based interfaces and cloud services. Cloud-based AI services (e.g., from OpenAI) are increasingly accepted as they adopt European data protection standards. Local hosting solutions were previously considered to ensure compliance with legal confidentiality requirements.

### Stage ###
Research

### Challenges ###
- Lack of centralized tool rollout across the firm
- Many tools still in trial or evaluation phase
- Legal and data protection concerns, especially around confidential client data
- System accreditation bottlenecks for cloud-based tools
- Tools often fail in complex or cross-border legal scenarios
- Limited benefit over manual research due to insufficient accuracy
- Internal resistance and delayed strategic alignment on AI adoption
- Difficulty integrating tools that require access to internal data repositories

### Source ###
Interview conducted in January 2025, led by sebis researchers. The interviewee is a senior legal professional, working at the organization.

# Legal AI Usage at a German E-Commerce Company

-------------

### Who is using it? ###
Legal operations and strategy teams within the corporate governance function are exploring and managing AI integration. Project managers and product leads, including non-technical staff, are involved in implementation.

### What problem(s) are they trying to solve? ###
- Improve efficiency in legal processes such as contract management and compliance.
- Enable self-service for frequently used documents like NDAs.
- Enhance searchability and usability of internal knowledge bases.
- Identify high-impact AI use cases for internal legal workflows.
- Reduce manual effort and dependency on disparate systems.

### Which NLP technologies are they using? ###
- Google Gemini, particularly the Notebook-LM feature, for general productivity, text editing, and internal chatbot creation.
- Conversational search features in a contract management tool are under consideration but not yet implemented.
- AI-powered search and chatbot functions in ServiceNow were explored, including a compliance chatbot based on company policy documents.
- All mentioned tools are cloud-based.

### Stage ###
Research

### Challenges ###
- Frequent delays and quality issues with external vendors during tool implementation.
- AI features in tools like Sirion and ServiceNow not yet activated due to ongoing basic setup and maintenance.
- Difficulty in identifying real efficiency gains without comprehensive quantification.
- Need for wider AI literacy and engagement among legal staff to identify relevant use cases.
- Ensuring AI tools fit into the existing tech landscape without adding unnecessary complexity.

### Source ###
Interview conducted in February 2025, led by sebis researchers. The interviewee is a project lead working at the organization.

Legal AI Usage at a German E-Commerce Company

# Legal AI Usage at German Law Firm

-------------

### Who is using it? ###
Legal AI tools are used by lawyers across multiple departments at the firm, including M&A, corporate law, tax, litigation, and investigations. A dedicated legal innovation team coordinates the introduction and support of these tools.

### What problem(s) are they trying to solve? ###
The primary goal is to process large volumes of legal documents more efficiently, especially in data-heavy fields such as M&A due diligence and litigation. Tasks include summarizing and analyzing documents, drafting emails or memos, automating repetitive tasks, and structuring legal facts. Legal AI is also being explored for translation services and everyday legal workflows like deadline tracking and email correspondence.

### Which NLP technologies are they using? ###
The firm uses general-purpose GenAI models such as ChatGPT and Harvey. Harvey was selected for its superior performance in legal contexts. Other tools tested include CoCounsel. These are cloud-based systems. Traditional contract analysis tools are also in use, but GenAI tools are increasingly preferred for their broader capabilities.

### Stage ###
Production

### Challenges ###
- Ensuring sustained and consistent tool usage across the firm
- Overcoming forgetfulness or lack of integration into daily habits
- Addressing skepticism or limited trust in AI-generated outputs
- Evaluating the real efficiency gains and quality improvements quantitatively
- Supporting users with training and prompting best practices
- Managing organizational change and user onboarding (Change Management)

### Source ###
Interview conducted in January 2025, led by sebis researchers. The interviewee is a legal innovation associate, working at the organization.


# Legal AI Usage at German Law Firm

-------------

### Who is using it? ###
Lawyers at a mid-sized law firm, including partners, associates, and legal departments such as litigation, insurance law, tax law, and antitrust.

### What problem(s) are they trying to solve? ###
They aim to streamline legal workflows, reduce manual labor, and improve efficiency in legal tasks such as document summarization, case assessment, and contract analysis. Use cases include litigation file analysis, initial assessments in insurance cases, tax case documentation, antitrust contract checks, and audit-related contract review.

### Which NLP technologies are they using? ###
They use large language models (LLMs) such as GPT-3.5, GPT-4, Claude, LLaMa, and Gemini. These are cloud-based but integrated with an internal anonymization pipeline to comply with data protection requirements. Each use case is linked with specific LLMs and system prompts optimized for the task.

### Stage ###
Production

### Challenges ###
- Low initial adoption despite positive reception
- Need for extensive user training and support
- Fear of AI disrupting the billable hour business model
- High effort for anonymizing and uploading internal datasets
- IT resource bottlenecks and long implementation timelines
- Varied quality and reliability of model outputs
- Difficulty in convincing senior staff of the tool’s benefits
- Complex coordination and prompt engineering for use case development

### Source ###
Interview conducted in January 2025, led by sebis researchers. The interviewee is a lawyer and partner, working at the organization.

# Legal AI Usage at German Law Firm

-------------

### Who is using it? ###
The primary users are legal professionals within a major German law firm, including lawyers and legal transformation teams. Legal engineers and AI experts are also involved in tool evaluation and integration.

### What problem(s) are they trying to solve? ###
The tools are used to support legal work through automation and enhanced document handling. Key applications include translation, improving document quality, contract analysis, clause evaluation, data extraction, and legal research. There is also interest in using AI for drafting support and comparing legal texts or emails. Retrieval-Augmented Generation (RAG) systems are used for precise answers based on curated legal literature.

### Which NLP technologies are they using? ###
They use general-purpose language models like ChatGPT, as well as AI services via Microsoft Azure (e.g., Copilot). Both cloud-based and on-premise systems are employed, including proprietary setups for secure document querying. Beta testing has been conducted with models accessing curated legal data.

### Stage ###
Production

### Challenges ###
- High barriers for tool approval due to strict internal standards and product vetting
- Tools often selected by tech-savvy staff, but underutilized by end-users (lawyers)
- Low user adoption and resistance due to unfamiliarity or unrealistic expectations
- Misunderstanding of AI limitations, especially regarding hallucinations and legal accuracy
- Difficulty curating and generalizing internal legal knowledge for AI use
- Lack of transparency in AI tools (“black box” issue) impedes legal reliability
- Structural gaps in legal education to prepare for AI-assisted work

### Source ###
Interview conducted in January 2025, led by sebis researchers. The interviewee is a legal professional, working at the organization.

# Legal AI Usage at Bavarian Ministry of Justice

-------------

### Who is using it? ###
Legal AI tools are used by a dedicated unit within the Bavarian Ministry of Justice focused on Legal Tech and AI. The tools are intended for use by courts and public prosecutors across Bavaria. Internal staff, including legal professionals and administrative personnel, are involved in the development and testing of these tools.

### What problem(s) are they trying to solve? ###
The Ministry is addressing challenges related to the increasing volume of documents in civil and criminal law cases, such as long legal briefs and large-scale digital evidence (e.g., terabytes of data from searches). Key use cases include document search, metadata extraction, anonymization of legal texts, drafting support, and summarization of lengthy documents. Internally, AI is also used for improving text quality and creating visual materials for presentations.

### Which NLP technologies are they using? ###
They use and test generative AI models like large language models (e.g., ChatGPT-style tools) and have piloted a custom anonymization prototype based on a language model. Projects are developed in collaboration with academic partners and rely on open-source models trained on judicial data. Tools are generally self-hosted or developed in close cooperation with universities to comply with strict legal and data protection requirements.

### Stage ###
Research

### Challenges ###
- Handling massive volumes of digital data in legal proceedings
- Ensuring data quality and digital availability (e.g., OCR issues)
- Anonymizing personal data
- Strict regulatory constraints (e.g., storage location laws, AI Act obligations)
- Complexity of interdepartmental rollouts (e.g., electronic case files)
- Addressing societal trust and ensuring transparency in AI use
- Limitations of generative AI outputs, especially for visuals and bias
- Need for high AI literacy among staff to use tools critically

### Source ###
Interview conducted in February 2025, led by sebis researchers. The interviewee is a legal professional working at the Bavarian Ministry of Justice.

Legal AI Usage at Bavarian Ministry of Justice

# Legal AI Usage at German Law Firm

-------------

### Who is using it? ###
The users are primarily corporate lawyers within the M&A department, including partners and associates. Additional involvement comes from litigation teams in the U.S. and global practice groups across the firm.

### What problem(s) are they trying to solve? ###
They aim to increase efficiency and speed in legal processes, such as document review, legal research, translation, and transaction management. Tools are used for information extraction, document drafting, summarization, and improving access to legal knowledge. They also support due diligence and e-discovery processes.

### Which NLP technologies are they using? ###
The firm uses extractive and generative AI tools, including general-purpose models like ChatGPT, especially within integrated platforms such as Juris-KI. They have tested tools like Kira and are preparing to roll out Westlaw Co-Counsel in Germany. DeepL is used regularly for legal document translation. Tools are primarily cloud-based, with strict security evaluations before deployment.

### Stage ###
Production

### Challenges ###
- High internal IT security and data protection requirements
- Lack of standardization in client AI usage policies
- Limited implementation efforts hindered tool adoption (e.g., Kira)
- Frustration with insufficient out-of-the-box performance of some tools
- Transparency concerns about data handling in AI tools
- Manual effort needed to verify AI-generated outputs
- Restrictions on AI features in communication tools due to legal risks

### Source ###
Interview conducted in January 2025, led by sebis researchers. The interviewee is a partner and an associate in the legal department, working at the organization.

# Legal AI Usage at German MedTech Company

-------------

### Who is using it? ###
The legal department, particularly in the IT legal function, is experimenting with AI tools. A small internal pilot team, including legal and IT staff, has tested tools such as ChatGPT and Microsoft Copilot.

### What problem(s) are they trying to solve? ###
The team is exploring AI tools for various legal support tasks, including:
- Drafting emails, PowerPoint slides, and speeches.
- Generating and reviewing legal documents (e.g., NDAs, development contracts).
- Running legal research and answering compliance-related questions.
- Reviewing long legal texts like EU regulations.
- Building internal legal chatbots (e.g., compliance Q&A bots).

### Which NLP technologies are they using? ###
They are using GPT-based models, specifically ChatGPT (commercial version), Microsoft Azure OpenAI (GPT-4), and Microsoft Copilot. The models are hosted in the cloud. They have not yet tested domain-specific tools like Harvey AI due to cost and availability concerns.

### Stage ###
Research

### Challenges ###
- High variability in output quality, especially for legal drafting and review.
- Tools fail to detect unusual or problematic contract clauses in longer documents.
- Inconsistent performance between tools (e.g., Azure OpenAI vs. Copilot).
- Legal hallucinations and unreliable responses in complex legal queries.
- Lack of efficiency gains for standard legal tasks.
- Data confidentiality and licensing concerns.
- Agents show promise but are still in early-stage testing.
- Legal risk tolerance is high as partial accuracy is insufficient in critical use cases.

### Source ###
Interview conducted in January 2025, led by sebis researchers. The interviewee is a legal counsel in IT law, working at the organization.

Legal AI Usage at German MedTech Company

# Legal AI Usage at German Law Firm

-------------

### Who is using it? ###
The primary users are legal professionals, including lawyers and legal departments within a large law firm affiliated with a major consulting organization. The interviewee oversees all AI-related initiatives at the firm.

### What problem(s) are they trying to solve? ###
They aim to streamline both legal and non-legal workflows. Use cases include document analysis (e.g., contract review), searching and extracting clauses, improving internal efficiency, accelerating legal assessments (e.g., compliance), and handling large-scale litigation by navigating massive document collections. The tools are also used for tasks like generating offers, board reports, and verifying NDAs or codes of conduct.

### Which NLP technologies are they using? ###
They use general-purpose language models such as ChatGPT and GPT-based systems. Tools are a mix of proprietary and customized applications, built on cloud platforms like Microsoft. Their solutions include:
- A customizable chatbot interface with agent chaining
- A self-developed contract assessment tool using retrieval methods and generative AI
- An AI agent framework for orchestrated document analysis in investigations and litigation

### Stage ###
Production

### Challenges ###
- Low initial user acceptance and difficulty encouraging engagement
- Need for highly tailored internal training to match specific legal fields
- Fragmented internal knowledge bases across legal teams
- Organizational resistance to new workflows
- Technical vendors lacking legal domain understanding
- Difficulty in knowledge curation for effective RAG systems

### Source ###
Interview conducted in January 2025, led by sebis researchers. The interviewee is a legal professional with a technical background, working at the organization.


# Legal AI Usage at Global Sportswear Company

-------------

### Who is using it? ###
In-house legal professionals at the corporate headquarters, particularly those supporting global IT, e-commerce, loyalty programs, data analytics, and AI compliance.

### What problem(s) are they trying to solve? ###
They aim to streamline legal workflows such as contract generation and review, translation, drafting non-standard contract templates, and email composition. Additionally, AI is used for summarizing meetings and providing structured outputs from long discussions. A future tool is expected to extract contract metadata to populate a database.

### Which NLP technologies are they using? ###
They are testing general-purpose cloud-based models such as Microsoft Co-Pilot (integrated with Microsoft 365) and a legal chatbot from beck-online. They are also exploring a contract software tool with AI-based data extraction capabilities.

### Stage ###
Research

### Challenges ###
- Limited availability of permanent licenses for tested tools
- Legal chatbot only supports certain areas of law and is language-specific
- AI outputs require verification to avoid hallucinations
- Users must be trained to critically evaluate AI responses
- Prompt quality significantly affects output usefulness
- High expectations can lead to frustration due to technical limitations
- Energy/resource usage concerns tied to widespread AI deployment

### Source ###
Interview conducted in February 2025, led by sebis researchers. The interviewee is an in-house legal professional, working at the organization.

Legal AI Usage at Global Sportswear Company

# Legal AI Usage at Mid-Sized German Law Firm

-------------

### Who is using it? ###
Lawyers and partners at a mid-sized law firm, especially those involved in corporate law and M&A. A small internal AI team, including a few partners and employees, is responsible for tool evaluation and testing.

### What problem(s) are they trying to solve? ###
The firm uses legal AI to support contract drafting, markup review, document screening for due diligence, summarization, legal research, and translation. AI also serves as a sparring partner in negotiations and helps structure legal arguments. Another goal is to reduce administrative work, such as client onboarding.

### Which NLP technologies are they using? ###
The firm uses ChatGPT (free and licensed versions) and Claude via third-party tools for general legal language tasks. Tools like Libra, LangDoc, and Spellbook are in testing. These rely on cloud-based large language models.

### Stage ###
Research

### Challenges ###
- Lack of technical infrastructure to ensure data privacy (e.g., anonymization, document redaction)
- Risk of hallucinations and factual inaccuracies in AI outputs, especially for legal norms and case law
- Limited internal expertise and resources for tool development or integration
- Low adoption due to unfamiliarity or resistance among employees
- Difficulty verifying AI-generated results without sufficient legal experience
- Restricted access to legal databases limits AI effectiveness
- Time constraints hinder extensive tool evaluation
- Fragmented tool landscape with varying stability and usefulness

### Source ###
Interview conducted in February 2025, led by sebis researchers. The interviewee is a partner and legal professional, working at the organization.


Legal AI Usage at Mid-Sized German Law Firm

# Legal AI Usage at a German Legal Insurance Subsidiary

-------------

### Who is using it? ###
Employees of a legal insurance subsidiary are involved in evaluating and implementing AI use cases. Legal services are primarily delivered through cooperation with external law firms and partners. Internal users are mostly part of service and claims departments.

### What problem(s) are they trying to solve? ###
The organization aims to automate legal claim evaluations, contract coverage checks, and reimbursement decisions (e.g., court or attorney costs). Use cases include extracting key information from legal documents, supporting legal research, and enabling more efficient internal processes. AI is also used indirectly through partner law firms for tasks like legal document drafting and analysis.

### Which NLP technologies are they using? ###
They use generative language models and prediction models, often as cloud-based licensed solutions. The models are integrated into broader systems for document processing and legal evaluation. There is also experimentation with on-premise and partially customized tools. External law firms employ tools based on large language models such as ChatGPT or GPT-4.

### Stage ###
Research

### Challenges ###
- Hallucinations and lack of precision in legal contexts
- Legal restrictions on providing legal advice internally
- Data protection concerns and GDPR-related limitations
- Difficulty in verifying outputs, especially for junior staff
- Complex integration with legacy systems
- High upfront effort for anonymization and data preparation
- Lack of suitable tools for civil-law (non-English) jurisdictions
- Risk of decreasing legal training quality among new professionals

### Source ###
Interview conducted in February 2025, led by sebis researchers. The interviewee is a managing director working at the organization.

Legal AI Usage at a German Legal Insurance Subsidiary

# Legal AI Usage at German Law Firm

-------------

### Who is using it? ###
Legal Tech practitioners within the law firm, including attorneys who integrate legal technology into client matters. A dedicated Legal Tech team coordinates tool procurement, implementation, and internal/external innovation processes.

### What problem(s) are they trying to solve? ###
The firm uses legal AI to improve efficiency in areas such as document review (e.g., eDiscovery and M&A due diligence), contract analysis, drafting, summarization of court decisions, and translation. AI also supports internal collaboration, document automation, and the creation of chronologies for fact-finding. The overarching goal is to enhance productivity and maintain competitiveness while meeting client expectations for efficient service.

### Which NLP technologies are they using? ###
They use both commercial and self-developed tools. General-purpose LLMs (e.g., GPT-based systems like Harvey) are used for legal content generation and review. Their custom tool is built on Azure services and performs document processing, search, and RAG-based QA. The models are hosted in the EU and designed without continuous learning, ensuring confidentiality. Other machine learning-based tools are also employed for M&A and document classification tasks.

### Stage ###
Production

### Challenges ###
- Compliance with strict German professional secrecy laws, requiring extensive vendor obligations.
- Internal coordination and documentation for custom tools are resource-intensive.
- LLMs struggle with deeper context understanding, document structure, and logical coherence.
- Identifying and communicating effective use cases across the organization remains challenging.

### Source ###
Interview conducted in January 2025, led by sebis researchers. The interviewee is a Legal Tech Manager and practicing attorney, working at the organization.

# Legal AI Usage at German Law Firm

-------------

### Who is using it? ###
A legal tech and digital transformation unit within a large law firm, consisting of jurists and IT-affine staff, is leading the AI implementation. The team includes one programmer and works closely with selected external service providers.

### What problem(s) are they trying to solve? ###
They aim to address repetitive legal workflows, quantitative document analysis, and efficiency improvements. Use cases include automation of contract generation, processing of routine administrative tasks, and document comparison. The overarching goals are reducing workload, minimizing errors, and driving innovation.

### Which NLP technologies are they using? ###
They are testing and integrating external and local large language models, including LLaMa, via APIs. Machine learning is used for quantitative analysis, and Robotic Process Automation (RPA) is also being implemented. Hybrid approaches combining ML and RPA are under development. These tools are introduced in a technology-neutral and problem-driven manner.

### Stage ###
Research

### Challenges ###
- High resource demands: time, cost, and organizational change
- Need for internal training and cultural acceptance
- Complexity of integrating multiple technologies into existing infrastructure
- Resistance to adoption without clear communication of benefits
- Lack of prompting methodology in early tool rollouts
- Vendor overselling during AI hype phases
- Regulatory and compliance concerns depending on tool use

### Source ###
Interview conducted in February 2025, led by sebis researchers. The interviewee is the head of digital transformation and legal tech, working at the organization.

# Legal AI Usage at German Law Firm

-------------

### Who is using it? ###
Lawyers across all levels (from associates to partners) and staff members, especially assistants, are using legal AI tools. A dedicated Legal Tech team, including in-house developers, supports implementation and customization.

### What problem(s) are they trying to solve? ###
The firm uses legal AI primarily to summarize legal documents, assist with drafting emails and first drafts of pleadings, and perform document analysis. They aim to reduce administrative burdens, improve efficiency, support knowledge management, and offer clients AI-enhanced legal services. Future plans include integrating internal systems for unified access and creating tools tailored to the firm's specific legal workflows.

### Which NLP technologies are they using? ###
They are using legal and general-purpose large language models, including Harvey, ChatGPT and Microsoft Copilot. Some tools are hosted in the cloud with data segregation ensured through contractual safeguards. Internal prototypes leverage LLMs, and software development is accelerated using AI-assisted coding tools like GitHub Copilot.

### Stage ###
Production

### Challenges ###
- Limited access to legal databases due to restrictive licensing from publishers.
- Initial skepticism and resistance from lawyers unfamiliar with AI capabilities.
- Lack of creativity among lawyers in identifying potential use cases.
- Significant variation in internal workflows makes standard tool adoption difficult.
- Technical limitations in Microsoft Copilot’s integration with existing infrastructure.
- Difficulty validating references to legal precedents using current tools.
- High effort required to ensure legal AI outputs meet quality standards.
- Resource overhead and risk of knowledge leakage when outsourcing tool development.

### Source ###
Interview conducted in February 2025, led by sebis researchers. The interviewee is a Director of Legal Tech, working at the organization.


# Legal AI Usage at German Law Firm

-------------

### Who is using it? ###
The tool is used by legal professionals including lawyers, business support staff, marketing, HR, and secretarial teams at a large international law firm. Usage is voluntary and access is available firm-wide to those who request it.

### What problem(s) are they trying to solve? ###
The tool is primarily used for drafting and refining legal clauses, generating legal memos, and formulating native-level English text (e.g., emails or documents). It helps with clause modifications, inspirations for restructuring contracts, and abstract legal analysis. In specific cases, it is also used to explore contract risks and options in restructuring scenarios. Beyond legal staff, non-legal teams use it for text generation tasks.

### Which NLP technologies are they using? ###
They are using Harvey, a proprietary chatbot interface based on large language models, trained with legal data. It allows document upload and context-aware interactions. Comparisons were made with ChatGPT and Microsoft Copilot, which is also being introduced. Usage is cloud-based but with contractual attention to data privacy and confidentiality.

### Stage ###
Production

### Challenges ###
- Data privacy and confidentiality concerns, especially under legal professional duties across multiple jurisdictions.
- Limited reliability when analyzing or generating content based on uploaded documents.
- Cannot yet draft complete contracts from term sheets.
- Quality limitations with German legal language and jurisdiction-specific clauses.
- Users must verify all AI outputs, as results may be confidently incorrect.
- Legal staff must be involved in tool implementation to ensure usability.
- Limited integration compared to tools embedded directly in office software (e.g., Copilot in Word/Outlook).

### Source ###
Interview conducted in December 2024, led by sebis researchers. The interviewee is a legal professional, working at the organization.

# Legal AI Usage at German Law Firm

-------------

### Who is using it? ###
Lawyers in the corporate law department, particularly those involved in M&A and venture capital transactions, are exploring and using legal AI tools. A dedicated internal AI team (approximately 8–9 people) supports these efforts.

### What problem(s) are they trying to solve? ###
The main goals are to improve efficiency in legal workflows, particularly in translation, document review (e.g., due diligence), summarization, and internal research. AI is also used to assist with tasks such as preparing presentations and navigating legal databases more efficiently.

### Which NLP technologies are they using? ###
They use general-purpose NLP models like ChatGPT and GPT-based tools. Tools are both cloud-based and on-premise. ChatGPT is used privately for general legal brainstorming, while Co-Pilot is tested for tasks like presentation and spreadsheet support. Tools such as DeepL are used extensively for document translation. Other tools mentioned include those for legal document extraction and comparison.

### Stage ###
Research

### Challenges ###
- Data privacy and confidentiality concerns
- High tool acquisition and implementation costs
- User adoption varies by personality, experience, and willingness to change
- Difficulty in integrating tools into the daily workflow
- Need for internal testing capacity and time
- Quality control, especially in consistent legal terminology during translation
- Risk of hallucinated results from some models (e.g., fictitious case law citations)

### Source ###
Interview conducted in February 2025, led by sebis researchers. The interviewee is a legal professional working at the organization.

# Legal AI Usage in a Corporate Legal Department

-------------

### Who is using it? ###
In-house legal counsel specializing in digital law, including IT procurement, data protection, and compliance with the EU AI Act. The tools are primarily used by legal professionals within the corporate legal department.

### What problem(s) are they trying to solve? ###
Legal AI tools are used to improve efficiency in legal research, document drafting, summarization of lengthy communication threads, and the preparation of presentations. Specific applications include drafting contracts, creating summaries from email histories, and assisting with internal compliance documentation related to data protection and the AI Act.

### Which NLP technologies are they using? ###
They use a mix of general-purpose and legal-specific LLM tools. These include:
- A chat function integrated into a legal database for research purposes.
- Microsoft Copilot for general-purpose AI tasks like text generation and summarization within the Office suite.
- A legal-specific LLM trained on European and German court decisions and some legal commentaries, used particularly for data protection and IT law.

All tools mentioned are cloud-based. ChatGPT is not officially approved for internal use, though it may be used privately or for non-confidential queries.

### Stage ###
Production

### Challenges ###
- Incomplete or imprecise responses from legal database chat tools, especially outside of their core legal domains.
- Difficulty accessing specific court decisions via the research chat interface.
- Need for high-quality prompt engineering to ensure accurate outputs.
- Ongoing requirement to manually verify AI-generated content.
- Restrictions on feeding personal or sensitive data into tools due to data protection concerns.
- Legal and organizational scrutiny around vendor selection and data residency (e.g., avoiding data transfer outside the EU).
- Limited internal authority of legal users over procurement and deployment decisions.

### Source ###
Interview conducted in February 2025, led by sebis researchers. The interviewee is a legal professional, working at the organization.

Legal AI Usage in a Corporate Legal Department

Syntactic analysis, also known as parsing, involves analyzing the grammatical structure of sentences to understand relationships between words. The fundamental purpose of syntactic analysis is to derive meaning from a text. In the analysis, checks are in place so that texts which are not meaningful or which do not make sense can be rejected. The analysis is done at an in-depth level starting from the basic symbols to all the way to an entire sentence. The concept is of critical importance since it enables machines to understand human language. It plays an instrumental role in bridging the gap between humans and systems. Example NLP techniques for syntactic analysis include dependency parsing, construction of a syntax tree, or part-of-speech tagging.

Syntactic Analysis

Application programming interface (API) recommendation plays a crucial role in facilitating smart contract development by providing developers with a ranked list of candidate APIs for specific recommendation points. Deep learning-based approaches have shown promising results in this field. However, existing approaches mainly rely on token sequences or abstract syntax trees (ASTs) for learning recommendation point-related features, which may overlook the essential knowledge implied in the relations between or within statements and may include task-irrelevant components during feature learning. To address these limitations, we propose a novel code graph called pruned and augmented AST (pa-AST). Our approach enhances the AST by incorporating additional knowledge derived from the control and data flow relations between and within statements in the smart contract code. Through this augmentation, the pa-AST can better represent the semantic features of the code. Furthermore, we conduct AST pruning to eliminate task-irrelevant components based on the identified flow relations. This step helps mitigate the interference caused by these irrelevant parts during the model feature learning process. Additionally, we extract the API sequence surrounding the recommendation point to provide supplementary knowledge for the model learning. The experimental results demonstrate our proposed approach achieving an average mean reciprocal rank (MRR) of 68.02%, outperforming the baselines' performance. Furthermore, through ablation experiments, we explore the effectiveness of our proposed code representation approach. The results indicate that combining pa-AST with the API sequence yields improved performance compared with using them individually. Moreover, our AST augmentation and pruning techniques significantly contribute to the overall results. © 2024 John Wiley & Sons Ltd.

Application programming interface recommendation for smart contract using deep learning from augmented code representation

This work introduces an innovative system tailored for automated document processing and advanced question-answering. The dataset creation code employs PyMuPDF and NL TK, orchestrating a systematic extraction of textual information from PDF files. This methodical process extends to content tokenization, where the text is segmented into structured sentences, culminating in the creation of a Pandas data Frame. The resultant dataset is then stored in the Parquet format, ensuring both efficient retrieval and in-depth analysis capabilities. The chatbot component of the system integrates ChainLit and Haystack libraries, incorporating a BM25 retriever for precise document retrieval and a generative language model for nuanced natural language interactions. Proficient in handling user queries, the chatbot adeptly retrieves relevant documents from the meticulously curated dataset and generates responses marked by conciseness and contextual accuracy. This paper not only delineates the architecture and methodologies employed in the system but also explores the diverse applications of this integrated approach. Representing a significant advancement in intelligent document-based inquiries, this system showcases versatility across a spectrum of fields, underlining its potential impact in research, customer support, and various domains relying on automated information retrieval and interpretation. © 2024 IEEE.

Mining Mate: A Chat Bot for Navigating Mining Regulations Using LLM Models

With the rapid development of my country’s socialist market economy, the system of joint and several liability has been established in my country’s civil and commercial law and is playing an increasingly important role. There are also problems such as scattered regulations and contradictory laws and regulations at the level. Since there is no unified application principle established in judicial practice, the litigation burden caused by the recovery lawsuit also wastes a lot of trial resources. Dimensional key features distinguish confusing charges. Use regular expression technology to extract key content such as fact descriptions, defendants’ charges, relevant laws and regulations in legal documents and create JSON format documents; use stammer word segmentation and stop word list to remove stop words; use Word2Vec algorithm to represent text into vector form, establish a judicial judgment prediction model and an optimization model, and through experimental comparison, it is concluded that the performance of the model after adding focal loss is improved by 1.82%, 0.45%, 1.62%, and 1.62% compared with the cross entropy loss, and the final accuracy of the optimized model is 84.78%., the precision rate is 87%, the recall rate is 85%, and the F1 value is 85%. The system is expected to assist judicial workers in classifying crimes with joint liability and reduce the burden of judicial workers reading many legal documents to classify crimes. © 2024 the Author(s).

Problems and Effective Countermeasures in Joint and Several Liability of Civil and Commercial Law Based on Deep Learning Assessment

The study addresses the possibilities of using information technology and natural language in the study of legal norms. The study aims to develop methods for using artificial intelligence and natural language processing to analyse jurisprudence. To achieve this goal, automatic strategies were created to recognise the main topics in legal texts, identify key legal concepts and analyse the structure of documents. The results of the study include an analysis of existing methods of using technology and natural language to analyse legal norms. The methods used included machine and deep learning, syntactic and semantic analysis, an automated classification system, relative analytics, and decision prediction. In addition, new methods of analysing legal texts based on artificial intelligence and natural language processing were introduced. These methods include the use of a thematic model that automatically identifies the main themes in legal texts, as well as automatic detection of legal concepts, which identifies key concepts. In addition, neural networks are used to analyse the structure of legal documents, which allows for more accurate recognition and analysis of various structural elements in documents. Automatic text generation based on legal information and ways to classify legal texts was also introduced. Thus, the main results are the automation of the process of analysing and understanding legal texts, an increase in the efficiency and accuracy of identifying thematic patterns and key legal concepts, and improved accessibility and speed of legal information processing. The results obtained indicate a great potential for the use of technological tools in jurisprudence, which can significantly improve the quality and accessibility of legal services, contributing to more efficient resolution of legal issues. © The Author(s). This is an open access article distributed under the terms of the Creative Commons Attribution License 4.0 (https://creativecommons.org/licenses/by/4.0/).

Possibilities of using artificial intelligence and natural language processing to analyse legal norms and interpret them

The utilization of machine learning methods for the analysis and interpretation of legal documents has been growing over the years, yet their potential and limitations remain under-explored. This study aims to address this gap, using unsupervised machine learning techniques to discover legal concepts from a corpus of Spanish legal documents. In addition to striving for optimal results, our research also embarks on an exploration of the challenges and limitations of unsupervised machine learning, investigating its capabilities and limitations in legal text analysis. We demonstrate that even relatively simplistic methodologies can yield noteworthy insights, with the highest identification rate of 70% achieved by Topic Modeling with Latent Dirichlet Allocation (LDA). However, challenges were encountered with the identification of some concepts, suggesting potential improvements in the corpus preprocessing and tokenization stages or the techniques to be used. The findings underscore the potential of unsupervised learning algorithms in legal text analysis, offering an intriguing path for future research. While acknowledging the need for higher accuracy in practical applications, this study emphasizes the remarkable feat achieved and proposes a way forward for a hybrid or adaptable approach. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024.

Exploring the Challenges and Limitations of Unsupervised Machine Learning Approaches in Legal Concepts Discovery

Judgments concerning animals have arisen across a variety of established practice areas. There is, however, no publicly available repository of judgments concerning the emerging practice area of animal protection law. This has hindered the identification of individual animal protection law judgments and comprehension of the scale of animal protection law made by courts. Thus, we detail the creation of an initial animal protection law repository using natural language processing and machine learning techniques. This involved domain expert classification of 500 judgments according to whether or not they were concerned with animal protection law. 400 of these judgments were used to train various models, each of which was used to predict the classification of the remaining 100 judgments. The predictions of each model were superior to a baseline measure intended to mimic current searching practice, with the best performing model being a support vector machine (SVM) approach that classified judgments according to term frequency—inverse document frequency (TF-IDF) values. Investigation of this model consisted of considering its most influential features and conducting an error analysis of all incorrectly predicted judgments. This showed the features indicative of animal protection law judgments to include terms such as ‘welfare’, ‘hunt’ and ‘cull’, and that incorrectly predicted judgments were often deemed marginal decisions by the domain expert. The TF-IDF SVM was then used to classify non-labelled judgments, resulting in an initial animal protection law repository. Inspection of this repository suggested that there were 175 animal protection judgments between January 2000 and December 2020 from the Privy Council, House of Lords, Supreme Court and upper England and Wales courts. © 2022, The Author(s).

Using machine learning to create a repository of judgments concerning a new practice area: a case study in animal protection law

Ethereum is a blockchain platform that allows developers to create smart contracts. Smart contracts are programs that can automatically execute and handle cryptocurrency funds. However, over a hundred thousand new smart contracts are deployed every day and inevitably contain vulnerabilities due to programming errors. Once deployed, smart contracts cannot be fixed or changed, leaving funds at risk. To mitigate it, we use deep learning to detect vulnerabilities in smart contracts. First, we create our own dataset of labeled smart contracts based on opcode sequences, since few smart contract codes and labeled datasets are publicly available. We collect opcode sequences by replaying real-world transactions from the Ethereum Mainnet in our fully synchronized node while we leverage a plugin called SODA"to label opcode sequences with vulnerability classes. Second, after data collection, we preprocess the data by removing duplicate opcode sequences, normalizing the sequences to the same length, and converting them into vectors. Finally, to detect vulnerabilities in smart contracts, we train a deep classification model using LSTM neural networks. Our model achieved an average accuracy of 82.63% and an F1-score of 79.74% across seven types of vulnerabilities, which is important for securing funds and logic in smart contracts. © 2023 IEEE."

Opcode Sequences-Based Smart Contract Vulnerabilities Detection Using Deep Learning

The acquisition of training data for AI modeling, especially in Natural Language Processing (NLP) and Question Answering (QA), is well-known as laborious and time-consuming process. However, the accessibility of such datasets is typically limited to specific domains and languages. In this paper, we introduce a framework designed for generating training data for such scenario, especially in the Vietnamese language and legal domain. Our approach only requires a source of document as input data, then it can generate a high quality dataset that can be used for model training from that data. We first extract named entities (NE) and part-of-speech (POS) components from these contexts, serves them as the basis for the QA pairs dataset. Subsequently, we transform these contexts into fill-in-the-blank (cloze) question formats and serve them as natural questions. Combining the answer and question pairs, we then construct a synthetic dataset. Finally, a simple filter was applied to eliminate the noise data from the synthetic dataset. In our experiments, the model using our method achieved a state-of-the-art performance on ALQAC dataset, a 68.52 in F1 score and 39.80 on exact match (EM) score. © 2023 IEEE.

VietLegalQA: Unsupervised Legal Question Answering for Vietnamese Using Cloze Translation Approach

In view of the existing traditional legal service, practitioners are unable to meet the huge demand; a large number of citizens are unable to determine the scope of the problems when they encounter infringement or require various legal assistance. Based on this, an automatic classification model of legal consultation based on Deep Learning is proposed in this paper. A KP+BiLSTM+Attention model is proposed. The Keyword Parser is introduced to extract key information. TF-IDF and part of speech tagging are used to filter out the important information in the user's legal problem description. The extracted keywords are given a weight value, and the other information weights are set to zero. The text information is transferred into two parallel word vector embedding layers. One of the word vector embedding layers transfers the results to the fusion layer for splicing, difference and point multiplication after the key information is converted into vector form. The output results are respectively connected with the results obtained from the other embedding layer as residuals. The final results are transferred to the BiLSTM+Attention model for training. The test results show that KP+BiLSTM+Attention model has significantly improved the accuracy and F1 value of the best benchmark method for text classification tasks of legal consulting. Therefore, KP+BiLSTM+Attention method has better performance in dealing with the classification of legal consulting issues. © 2023, Science and Information Organization. All Rights Reserved.

Research on the Text Classification of Legal Consultation Based on Deep Learning

The proposed work aims to develop a legal document summarization tool using abstractive summarization techniques, with a focus on creating informative and concise summaries of legal documents. Modern deep neural network based pretrained models such as Distilled Bi-directional and Auto-regressive Transformer(DistillBart), Legal Pegasus, and Legal Longformer Encoder-Decoder(LED) are used to deal with the problem of low training dataset availability. Various sentence ranking approaches such as keywords, Parts Of Speech(POS) tagging, chunking based approach, and the position of sentences in the document and sentence centrality based on similarity are used to deal with the issue of large size of legal judgements. To improve informativeness of summary, focus is on entities such as important dates, concerned people, and referred acts in the legal case. The performance of different models employed for summarization are evaluated using evaluation metrics such as Bilingual Evaluation Understudy(BLEU), Precision, Recall, F1 Score, and Recall Oriented Understudy for Gisting Evaluation(ROUGE). The proposed system is a valuable tool for legal professionals to quickly understand the main points of a legal document and make informed decisions.  © 2023 IEEE.

Legal Data Assistive Tool Using Deep-Learning

Recently, security issues of smart contracts are arising great attention due to the enormous financial loss caused by vulnerability attacks. There is an increasing need to detect similar codes for hunting vulnerability with the increase of critical security issues in smart contracts. Binary similarity detection that quantitatively measures the given code diffing has been widely adopted to facilitate critical security analysis. However, due to the difference between common programs and smart contract, such as diversity of bytecode generation and highly code homogeneity, directly adopting existing graph matching and machine learning based techniques to smart contracts suffers from low accuracy, poor scalability and the limitation of binary similarity on function level. Therefore, this paper investigates graph neural network to detect smart contract binary code similarity at the program level, where we conduct instruction-level normalization to reduce the noise code for smart contract pre-processing and construct contract control flow graphs to represent smart contracts. In particular, two improved Graph Convolutional Network (GCN) and Message Passing Neural Network (MPNN) models are explored to encode the contract graphs into quantitatively vectors, which can capture the semantic information and the program-wide control flow information with temporal orders. Then we can efficiently accomplish the similarity detection by measuring the distance between two targeted contract embeddings. To evaluate the effectiveness and efficient of our proposed method, extensive experiments are performed on two real-world datasets, i.e., smart contracts from Ethereum and Enterprise Operation System (EOS) blockchain-based platforms. The results show that our proposed approach outperforms three state-of-the-art methods by a large margin, achieving a great improvement up to 6.1% and 17.06% in accuracy. © 2023 Tech Science Press. All rights reserved.

Deep Learning-Based Program-Wide Binary Code Similarity for Smart Contracts

Precise extraction of the required activities from contract documents is crucial to develop a complete and accurate construction schedule. Currently, the scheduler is required to read the complete lengthy contract package to extract activities and convert them into a structured format to prepare a schedule. The current manual methods of activities extraction for schedule development are inefficient, laborious, and time-consuming. Although a few researchers have developed models for automated information extraction from legal documents, those models are applicable to quantitative requirements only. To address this, the current study proposes an automated information retrieval system to extract the activities provided in contracts using natural language processing (NLP) methods. The study implemented the dependency parsing technique that employed the syntactic features of requirement text to extract the actors, actions, objects, and conditions specified in the requirement. The model achieved an average recall and precision of 94% and 95%, respectively, on the requirements of a real design-build project. The proposed automated activities retrieval system is expected to help the schedulers develop the construction schedules. © 2022 ASCE.

Extraction of Activities Information from Construction Contracts Using Natural Language Processing (NLP) Methods to Support Scheduling

Legal information processing attracts attention from a number of organizations globally which includes research institutions; specific areas of interest ranges from representation of legal data, such as prior court cases, for countries which adopt common law system. These legal datasets typically include legislative acts or statutes, which vary from one country to another. Mining this legal data in order to extract useful information poses formidable challenges, which include coming up with ways for storing datasets, which are continuously generated and growing exponentially every year. Additional challenges include keeping up with amendments of the statutes and invalidating old statutes. This paper details a system architecture containing several key subsystems toward design and development of Legal Humanities for Africa, which is digital, query-able and easy to use and navigate by both lay user and experienced professionals. The Legal Humanities of Africa architecture has three modules, namely, knowledge base, knowledge engine and HCI module. The knowledge base handles the legal data dictionary, glossary and metadata in a domain-specific manner, while the knowledge engine handles processing data from the legal cases or statutes. Various approaches to computational linguistics, such as natural language processing or information extraction, have been used for natural language processing, including finding part of speech in text that contains most informative terms that are useful in computing similarity between prior cases to user queries. Submodules in knowledge base are also employed as needed to help optimize the process of extracting useful information to users usually in terms of relevant cases relating to specific legal matters at hand. Other techniques employed include several aspects of machine learning, such as unsupervised learning approaches to identify and cluster prior cases so that similar cases are clustered together, thus, making the process of matching user queries to prior cases easy. The HCI module provides user-specific (lay vs. experts), domain-specific and other anchor desks as required in a typical large application that can be commonly used by many types of users. A parametric evaluation of the performance of the Cloud-based Legal Architecture indicates that the system can enhance its use by both professionals and commoners; details are provided in this paper. It is hoped that the Legal Humanities of Africa architecture will become the benchmark architecture for Africa at large. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

Deep Learning-Based Legal System Architecture for Africa: An Architectural Study

The nation's massive underground utility infrastructure must comply with a multitude of regulations. The regulatory compliance checking of underground utilities requires an objective and consistent interpretation of the regulations. However, utility regulations contain a variety of domain-specific terms and numerous spatial constraints regarding the location and clearance of underground utilities. It is challenging for the interpreters to understand both the domain and spatial semantics in utility regulations. To address the challenge, this paper adopts an ontology and rule-based Natural Language Processing (NLP) framework to automate the interpretation of utility regulations – the extraction of regulatory information and the subsequent transformation into logic clauses. Two new ontologies have been developed. The urban product ontology (UPO) is domain-specific to model domain concepts and capture domain semantics on top of heterogeneous terminologies in utility regulations. The spatial ontology (SO) consists of two layers of semantics – linguistic spatial expressions and formal spatial relations – for better understanding the spatial language in utility regulations. Pattern-matching rules defined on syntactic features (captured using common NLP techniques) and semantic features (captured using ontologies) were encoded for information extraction. The extracted information elements were then mapped to their semantic correspondences via ontologies and finally transformed into deontic logic (DL) clauses to achieve the semantic and logical formalization. The approach was tested on the spatial configuration-related requirements in utility accommodation policies. Results show it achieves a 98.2% precision and a 94.7% recall in information extraction, a 94.4% precision and a 90.1% recall in semantic formalization, and an 83% accuracy in logical formalization. © 2021 Elsevier Ltd

Ontology and rule-based natural language processing approach for interpreting textual regulations on underground utility infrastructure

The Legal Clinics (LC) are no-profit programs aimed to provide legal assistance to low-income citizens. Given the large number of consultations requested by the LC, the processes usually are delayed. This delay entails higher costs for the institutions that offer this legal advice. This paper explores the low-cost automation of legal problem identification in LC through the use of Natural Language Processing (NLP) techniques. We propose a methodology of text processing for the legal issue identification in the legal complaint assistance in LC. The method is based on preprocessing of the text, word to vector transformation, text identification models, and model evaluation for final classification. This methodology looks to accelerate the first step of the legal consultation process identifying the legal issue described by the LC user. The method is evaluated using real cases from the LC of the Santo Tomas University of Colombia. The results provided by the methodology depict a performance of around 95% for the legal issue identification. It is expected that this system will contribute to the delay decrease in the legal advice from LC, and it will help increase the number of advised users through virtual legal assistance. © 2021 The Authors. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/)

Automation system based on NLP for legal clinic assistance

Analysing judicial documents is a significant task in the field of legal intelligence, of which the prediction of law articles can help ordinary people better comprehend basic legal knowledge. It can be seen as a multi-label classification problem. Our work uses description of the cases as the input and views law articles as labels. After word segmentation and feature extraction, the classifier model TextCNN is used to train and predict the results. Results of comparative experiments show that our approach can achieve competitive results. © 2019 IEEE.

Law Article Prediction Based on Deep Learning

Legal texts express conditions in natural language describing what is permitted, forbidden or mandatory in the context they regulate. Despite the numerous approaches tackling the problem of moving from a natural language legal text to the respective set of machine-readable conditions, results are still unsatisfiable and it remains a major open challenge. In this paper, we propose a preliminary approach which combines different Natural Language Processing techniques towards the extraction of rules from legal documents. More precisely, we combine the linguistic information provided by WordNet together with a syntax-based extraction of rules from legal texts, and a logic-based extraction of dependencies between chunks of such texts. Such a combined approach leads to a powerful solution towards the extraction of machine-readable rules from legal documents. We evaluate the proposed approach over the Australian “Telecommunications consumer protections code”. © 2018, Springer Nature Switzerland AG.

Combining Natural Language Processing Approaches for Rule Extraction from Legal Documents

Query reformulation is the process of iteratively modifying a query to improve the quality of search engine results. In recent years, the task of reformulating natural language (NL) queries has received considerable diligence from both industry and academic communities. Traditionally, query reformulation has been mostly approached by using the noisy channel model. Since legal queries are diverse and multi-faceted, these traditional approaches cannot effectively handle low frequency and out-of-vocabulary (OOV) words. Motivated by these issues, we rethink the task of legal query reformulation as a type of monolingual neural machine translation (NMT) problem, where the input (source) query is potentially erroneous and the output (target) query is its corrected form. We propose a unified and principled framework with multiple levels of granularity. Specifically, (i) an encoder with character attention which augments the subword representation; (ii) a decoder with attentions that enable the representations from different levels of granularity to control the translation cooperatively and (iii) a semi-supervised methodology to extract and augment a large-scale dataset of NL query pairs combining syntactic and semantic operations. We establish the effectiveness of our methodology using an internal dataset, where the training data is automatically obtained from user query logs. We further demonstrate that training deep neural networks on additional data with synthesized errors can improve performance for translation. © 2019 Copyright held by the owner/author(s).

Legal query reformulation using deep learning

Currently, pest management practices require modern equipment and the use of complex information, such as regulations and guidelines. The complexity of regulations is the root cause of the emergence of automated solutions for compliance assessment by translating regulations into sets of machine-processable rules that can be run by specialized modules of farm management information systems (FMIS). However, the manual translation of rules is prohibitively costly, and therefore, this translation should be carried out with the support of artificial intelligence techniques. In this paper, we use the official Spanish phytosanitary products registry to empirically evaluate the performance of four popular machine learning algorithms in the task of correctly classifying pesticide regulations as prohibitions or obligations. Moreover, we also evaluate how to improve the performance of the algorithms in the preprocessing of the texts with natural language processing techniques. Finally, due to the specific characteristics of the texts found in pesticide regulations, resampling techniques are also evaluated. Experiments show that the combination of the machine learning algorithm Logic regression, the natural language technique part-of-speech tagging and the resampling technique Tomek links is the best performing approach, with an F1 score of 68.8%, a precision of 84.46% and a recall of 60%. The experimental results are promising, and they show that this approach can be applied to develop a computer-aided tool for transforming textual pesticide regulations into machine-processable rules. To the best of our knowledge, this is the first study that evaluates the use of artificial intelligence methods for the automatic translation of agricultural regulations into machine-processable representations. © 2018 Elsevier B.V.

Machine learning for automatic rule classification of agricultural regulations: A case study in Spain

In this work we present the TULSI system (so named after Turin University Legal Semantic Interpreter), a system to produce automatic annotations of normative documents through the extraction of modificatory provisions. TULSI relies on a deep syntactic analysis and a shallow semantic interpreter that are illustrated in detail. We report the results of an experimental evaluation of the system and discuss them, also suggesting future directions for further improvement. © 2012 Springer Science+Business Media B.V.

TULSI: An NLP system for extracting legal modificatory provisions

We describe a legal question answering system which combines legal information retrieval and textual entailment. We have evaluated our system using the data from the first competition on legal information extraction/ entailment (COLIEE) 2014. The competition focuses on two aspects of legal information processing related to answering yes/no questions from Japanese legal bar exams. The shared task consists of two phases: legal ad hoc information retrieval and textual entailment. The first phase requires the identification of Japan civil law articles relevant to a legal bar exam query. We have implemented two unsupervised baseline models (tf-idf and Latent Dirichlet Allocation (LDA)-based Information Retrieval (IR)), and a supervised model, Ranking SVM, for the task. The features of the model are a set of words, and scores of an article based on the corresponding baseline models. The results show that the Ranking SVM model nearly doubles the Mean Average Precision compared with both baseline models. The second phase is to answer “Yes” or “No” to previously unseen queries, by comparing the meanings of queries with relevant articles. The features used for phase two are syntactic/semantic similarities and identification of negation/antonym relations. The results show that our method, combined with rule-based model and the unsupervised model, outperforms the SVM-based supervised model. © Springer-Verlag Berlin Heidelberg 2015.

Legal question answering using ranking SVM and syntactic/semantic similarity

Previous work has shown that modeling relationships between articles of a regulation as vertices of a graph network works twice as better than traditional information retrieval systems for returning articles relevant to the question. In this work we experiment by using natural language techniques such as lemmatizing and using manual and automatic thesauri for improving question based document retrieval. For the construction of the graph, we follow the approach of representing the set of all the articles as a graph; the question is split in two parts, and each of them is added as part of the graph. Then several paths are constructed from part A of the question to part B, so that the shortest path contains the relevant articles to the question. We evaluate our method comparing the answers given by a traditional information retrieval system - vector space model adjusted for article retrieval, instead of document retrieval - and the answers to 21 questions given manually by the general lawyer of the National Polytechnic Institute, based on 25 different regulations (academy regulation, scholarships regulation, postgraduate studies regulation, etc.); with the answer of our system based on the same set of regulations. We found that lemmatizing increases performance in around 10%, while the use of thesaurus has a low impact. © Springer-Verlag Berlin Heidelberg 2009.

NLP for shallow question answering of legal documents using graphs

Ongoing research in information retrieval offers new approaches demonstrating that the traditional methods may have reached their limitations. In fact, actual results to date have shown that advanced systems will demonstrate effectiveness at combining multiple text and user's need representations. This paper suggests to supplement the traditional conceptual abstracts by a syntactical description designed to improve retrieval performance. This choice is carefully motivated and examples are given. Without attempting to reason about the documents content, a domain-specific knowledge representation is used to form clusters of concepts. The point is to represent the joint structure of the component-concepts involved in a composite-concept. This approach extends the reach of key concepts assignment to: interactive query elicitation; multiwords phrase search; and finer-grained thesaurus relationships. This paper emphasizes that intellectual assignment of features is part of artificial intelligence approach to information retrieval. The principles of implementation are also described.

Contribution to indexing in legal information retrieval

The complexity and diversity of government regulations make understanding the regulations a non-trivial task. One of the issues is the existence of multiple sources of regulations and interpretive guides; the latter are often independent of governing bodies. This work aims to develop an information infrastructure for legal information retrieval with applications to electronic-rulemaking. The pilot study focuses on accessibility regulations from the US Federal government, private organizations and European agencies. A shallow parser is developed to consolidate different regulations into a unified XML format, which is well suited for handling semi-structured data such as legal documents. Handcrafted rules and a text mining tool are developed to extract the important features, such as concepts, measurements, effective dates and so on, and to incorporate them into the corpus.To compare and locate related provisions from different regulatory documents, we employ Information Retrieval techniques to combine generic features with domain knowledge. Structural information from regulations, such as the hierarchical organization of provisions and heavy referencing among provisions, are used to help improve the relatedness analysis. Results are obtained to illustrate the use of regulatory structure and domain knowledge in provision comparisons. Application to an e-rulemaking scenario for a rights-of-way drafted regulation is shown to demonstrate extended capabilities of the prototype system. Copyright 2005 ACM.

Legal information retrieval and application to E-rulemaking

This paper introduces GENTRAC, an open-access web-based tool built to interactively detect and analyze potentially traumatic content in witness statements of genocide and mass atrocity trials. Harnessing recent developments in natural language processing (NLP) to detect trauma, GENTRAC processes and formats court transcripts for NLP analysis through a sophisticated parsing algorithm and detects the likelihood of traumatic content for each speaker segment. The tool visualizes the density of such content throughout a trial day and provides statistics on the overall amount of traumatic content and speaker distribution. Capable of processing transcripts from four prominent international criminal courts, including the International Criminal Court (ICC), GENTRAC's reach is vast, tailored to handle millions of pages of documents from past and future trials. Detecting potentially re-traumatizing examination methods can enhance the development of trauma-informed legal procedures. GENTRAC also serves as a reliable resource for legal, human rights, and other professionals, aiding their comprehension of mass atrocities' emotional toll on survivors.

GENTRAC: A Tool for Tracing Trauma in Genocide and Mass Atrocity Court Transcripts

This study aims to tackle some challenges posed by legal texts in the field of NLP. The LegalEval challenge proposes three tasks, based on Indial Legal documents: Rhetorical Roles Prediction, Legal Named Entity Recognition, and Court Judgement Prediction with Explanation. Our work focuses on the first two tasks. For the first task we present a context-aware approach to enhance sentence information. With the help of this approach, the classification model utilizing InLegalBert as a transformer achieved 81.12% Micro-F1. For the second task we present a NER approach to extract and classify entities like names of petitioner, respondent, court or statute of a given document. The model utilizing XLNet as transformer and a dependency parser on top achieved 87.43% Macro-F1.

TeamUnibo at SemEval-2023 Task 6: A transformer based approach to Rhetorical Roles prediction and NER in Legal Texts

We present GENTLE, a new mixed-genre English challenge corpus totaling 17K tokens and consisting of 8 unusual text types for out-of-domain evaluation: dictionary entries, esports commentaries, legal documents, medical notes, poetry, mathematical proofs, syllabuses, and threat letters. GENTLE is manually annotated for a variety of popular NLP tasks, including syntactic dependency parsing, entity recognition, coreference resolution, and discourse parsing. We evaluate state-of-the-art NLP systems on GENTLE and find severe degradation for at least some genres in their performance on all tasks, which indicates GENTLE's utility as an evaluation dataset for NLP systems.

GENTLE: A Genre-Diverse Multilayer Challenge Set for English NLP and Linguistic Evaluation

Legal-ES is an open source resource kit for legal Spanish. It consists of a large scale Spanish corpus of open legal texts and different kinds of language models including word embeddings and topic models. The corpus includes over 1000 million words covering a collection of legislative and administrative open access documents in Spanish from different sources representing international, national and regional entities. The corpus is pre-processed and tokenized using Spacy. For the word embeddings, gensim was used on the collection of tokens, producing a representation space that is especially suited to reflect the inherent characteristics of the legal domain. We calculate also topic models to obtain a convenient tool to understand the main topics in the corpus and to navigate through the documents exploiting the semantic similarity among documents. We will analyse the time structure of a dynamic topic model to infer changes in the legal production of Spanish jurisdiction that have occurred over the analysed time framework.

Legal-ES: A Set of Large Scale Resources for Spanish Legal Text Processing

In this work we present a corpus for the evaluation of sensitive information detection approaches that addresses the need for real world sensitive information for empirical studies. Our sentence corpus contains different notions of complex sensitive information that correspond to different aspects of concern in a current trial of the Monsanto company. This paper describes the annotations process, where we both employ human annotators and furthermore create automatically inferred labels regarding technical, legal and informal communication within and with employees of Monsanto, drawing on a classification of documents by lawyers involved in the Monsanto court case. We release corpus of high quality sentences and parse trees with these two types of labels on sentence level. We characterize the sensitive information via several representative sensitive information detection models, in particular both keyword-based (n-gram) approaches and recent deep learning models, namely, recurrent neural networks (LSTM) and recursive neural networks (RecNN). Data and code are made publicly available.

A Real-World Data Resource of Complex Sensitive Sentences Based on Documents from the Monsanto Trial

This article presents the current outcomes of the MARCELL CEF Telecom project aiming to collect and deeply annotate a large comparable corpus of legal documents. The MARCELL corpus includes 7 monolingual sub-corpora (Bulgarian, Croatian, Hungarian, Polish, Romanian, Slovak and Slovenian) containing the total body of respective national legislative documents. These sub-corpora are automatically sentence split, tokenized, lemmatized and morphologically and syntactically annotated. The monolingual sub-corpora are complemented by a thematically related parallel corpus (Croatian-English). The metadata and the annotations are uniformly provided for each language specific sub-corpus. Besides the standard morphosyntactic analysis plus named entity and dependency annotation, the corpus is enriched with the IATE and EUROVOC labels. The file format is CoNLL-U Plus Format, containing the ten columns specific to the CoNLL-U format and four extra columns specific to our corpora. The MARCELL corpora represents a rich and valuable source for further studies and developments in machine learning, cross-lingual terminological data extraction and classification.

The MARCELL Legislative Corpus

The complexity and diversity of government regulations make understanding the regulations a non-trivial task. One of the issues is the existence of multiple sources of regulations and interpretive guides; the latter are often independent of governing bodies. This work aims to develop an information infrastructure for legal information retrieval with applications to electronic-rulemaking. The pilot study focuses on accessibility regulations from the US Federal government, private organizations and European agencies. A shallow parser is developed to consolidate different regulations into a unified XML format, which is well suited for handling semi-structured data such as legal documents. Handcrafted rules and a text mining tool are developed to extract the important features, such as concepts, measurements, effective dates and so on, and to incorporate them into the corpus.To compare and locate related provisions from different regulatory documents, we employ Information Retrieval techniques to combine generic features with domain knowledge. Structural information from regulations, such as the hierarchical organization of provisions and heavy referencing among provisions, are used to help improve the relatedness analysis. Results are obtained to illustrate the use of regulatory structure and domain knowledge in provision comparisons. Application to an e-rulemaking scenario for a rights-of-way drafted regulation is shown to demonstrate extended capabilities of the prototype system.

Legal information retrieval and application to e-rulemaking

The present work proposes a legal assistance tool based on NLP and a predictive ML model adapted to the particularities of Colombian Justice system for the Legal Defense Agency of the Colombian State (ANDJE), It is developed in 4 parts associated with specific requirements: An Exploratory Data Analysis of the available information, an NLP pre-processing of large volumes of information structured in databases and not structured in scanned PDFs, development of predictive models and user assistance tool based in Transformers. Combining in an original and innovative way predictive, information presentation, similarity searching engine and summary, Question Answering modules for treatment of large legal texts, to increase the probability of success in defending the interests of the Colombian state. Considerations for future works.

Assistance System for Judicial Awards for The Colombian State Legal Defense Agency Through NLP -ANDJE-

Ongoing research in information retrieval offers new approaches, demonstrating that the traditional methods may have reached their limitations. In fact, actual results to date have shown that advanced systems will demonstrate effectiveness at combining multiple text and user need representations. The paper suggests supplementing the traditional conceptual abstracts with a syntactical description designed to improve retrieval performance. This choice is carefully motivated and examples are given. Without attempting to reason about the document contents, a domain specific knowledge representation is used to form clusters of concepts. The point is to represent the joint structure of the component concepts involved in a composite concept. This approach extends the reach of key concept assignment to: interactive query elicitation; multiword phrase search; finer grained thesaurus relationships. The paper emphasizes that intellectual assignment of features is part of an artificial intelligence approach to information retrieval. The principles of implementation are also described.

A contribution to indexing in legal information retrieval

This paper describes an information extraction system that identifies and analyzes statements in court opinions regarding the value of cited cases as precedents. The system employs partial parsing techniques in conjunction with a semantic grammar to identify the language associated with such rulings. The most novel aspect of the system lies in its anaphora resolution module that combines syntactic, semantic, and domain-specific inference rules with local discourse information to link such language to case references.

Anaphora resolution in the extraction of treatment history language from court opinions by partial parsing

Aide (`AustLII Inferencing Development Environment') provides a quasi-natural language form of knowledge representation which is reasonably close to statutory language, but at the same time represents knowledge so that it can be used by an inferencing engine using predicate calculus.Aide also provides a supportive development environment for the user by parse tree differentiation to assist the user choosing the correct parsing of rules. The development environment and the user environment are both web-based, facilitating collaborative development of knowledgebases, and the integration of inferencing dialogues with legal source texts on the web.

AustLll's aide — natural language legislative rulebases

The Dutch Tax and Customs Administration (DTCA) is one of many organizations that deal with a multitude of electronic legal data, from various sources and in different formats. In this paper, we describe the results of a study aimed at better access to these sources by having a supplier and format independent knowledge store that describes the sources and their interrelations in a semantic network. Furthermore we developed parsers to automatically detect the identity of sources and typed references within the sources to other legal documents. These parsers can be used to fill and update the semantic network as new documents are added.

Constructing a semantic network for legal content

Text classification is an important task in the legal domain. In fact, most of the legal information is stored as text in a quite unstructured format and it is important to be able to automatically classify these texts into a predefined set of concepts.Support Vector Machines (SVM), a machine learning algorithm, has shown to be a good classifier for text bases [12]. In this paper, SVMs are applied to the classification of European Portuguese legal texts - the Portuguese Attorney General's Office Decisions - and the relevance of linguistic information in this domain, namely lemmatisation and part-of-speech tags, is evaluated.The obtained results show that some linguistic information (namely, lemmatisation and the part-of-speech tags) can be successfully used to improve the classification results and, simultaneously, to decrease the number of features needed by the learning algorithm.

Is linguistic information relevant for the classification of legal texts?

One of the main obstacles to progress in the field of artificial intelligence and law is the natural language barrier, but the technology of natural language processing has advanced recently. In this paper, we will show that a state-of-the-art statistical parser can handle even the complex syntactic constructions of an appellate court judge, and that a deep semantic interpretation of the full text of a judicial opinion can be computed automatically from the output of the parser. Our ultimate goal is to use this semantic interpretation to extract from a judicial opinion precisely the information that a lawyer wants to know about a case.

Deep semantic interpretations of legal texts

The ultimate goal of the research line described here is support for automated modelling of sources of law. One of the first steps is the automatic recognition of norms. In earlier work we presented a categorization of norms or provisions in legislation. We claimed that the categories are characterized by the use of typical sentence structures and that this would enable automatic detection and classification. In this paper we present the results of experiments in such automatic classification of provisions. We have defined fourteen different categories of provisions, and compiled a list of 88 sentence structures for those categories from twenty Dutch laws. Based on these structures, a parser was used to classify the sentences in fifteen different Dutch laws, classifying 91% of 592 sentences correctly. It compares well with other, statistical approaches. An important improvement of our classifier will be the distinction of principal and auxiliary sentences.

A next step towards automated modelling of sources of law

In this paper we illustrare a research based on NLP techniques aimed at automatically annotate modificatory provisions. We propose an approach which pairs deep syntactic parsing with rule-based shallow semantic analysis relying on a fine-grained taxonomy of modificatory provisions. The implemented system is evaluated on a large dataset hand-crafted by legal experts; the results are discussed and future directions of the research outlined.

NLP-based extraction of modificatory provisions semantics

One open problem in the AI &amp; Law community is how to provide computers with a basic understanding of legal concepts, and their relationship with legal texts and with the legal lexicon. We propose to add a layer to connect the linguistic description of the provisions to syntactic patterns using FramNet that can be exploited thought NLP tools. A deep-parsing and shallow-semantics approach has been devised to interpret and retrieve the characterizing components of legal modificatory provisions. In this paper we single out the case of efficacy suspension and show how FrameNet approach can provide profit especially to isolate temporal parameters and their interpretation.

FrameNet model of the suspension of norms

In the last few years University of Turin and CIRSFID University of Bologna collaborated to pair NLP techniques and legal knowledge to detect modificatory provisions in normative texts. Annotating these modifications is a relevant and interesting problem, in that modifications affect the whole normative system; and legal language, though more regular than unrestricted language, is sometimes particularly convoluted, and poses specific linguistic issues. This paper focuses on two major aspects. First, we explore a combination between parsing and regular expressions; to the best of our knowledge, such hybrid strategy has never been proposed before to tackle the problem at hand. Secondly, we significantly extend past works coverage (basically focussed on substitution, integration and repeal modifications) in order to account for further twelve modification kinds. For the sake of conciseness, we fully illustrate and discuss only few modification types that are more relevant and interesting: suspension, prorogation of efficacy, postponement of efficacy and exception/derogation. These sorts of modifications appear particularly challenging, in that modifications in these categories make use of similar linguistic speech acts and verbs, and exhibit strong similarities in the linguistic syntactical patterns, to such an extent that to discern them is difficult for the legal expert, too. We describe the implemented system and report about an extensive experimentation on the new modificatory provisions. Results are discussed in order to improve both system's accuracy and annotation practice.

Modificatory provisions detection: a hybrid NLP approach

Improving the readability of legislation is an important and unresolved problem. Recently, researchers have begun to apply legal informatics to this problem. This paper applies machine learning to predict the readability of sentences from legislation and regulations. A corpus of sentences from the United States Code and US Code of Federal Regulations was created. Each sentence was labelled for language difficulty using results from a large-scale crowdsourced study undertaken during 2014. The corpus was used as training and test data for machine learning. The corpus includes a version tagged using the Stanford parser context free grammar and a version tagged using the Stanford dependency grammar parser. The corpus is described and made available to interested researchers. We investigated whether extending natural language features available as input to machine learning improves the accuracy of prediction. Among features evaluated are those from the context free and dependency grammars. Letter and word ngrams were also studied. We found the addition of such features improves accuracy of prediction on legal language. We also undertake a correlation study of natural language features and language difficulty drawing insights as to the characteristics that may make legal language more difficult. These insights, and those from machine learning, enable us to describe a system for reducing legal language difficulty and to identify a number of suggested heuristics for improving the writing of legislation and regulations.

Machine learning for readability of legislative sentences

We study how contract element extraction can be automated. We provide a labeled dataset with gold contract element annotations, along with an unlabeled dataset of contracts that can be used to pre-train word embeddings. Both datasets are provided in an encoded form to bypass privacy issues. We describe and experimentally compare several contract element extraction methods that use manually written rules and linear classifiers (logistic regression, SVMs) with hand-crafted features, word embeddings, and part-of-speech tag embeddings. The best results are obtained by a hybrid method that combines machine learning (with hand-crafted features and embeddings) and manually written post-processing rules.

Extracting contract elements

In this paper, we present a German supertagger that analyses syntactic functions in linear order. We apply a statistical sequential model, conditional random fields (CRF), to Swiss law texts, in a real world scenario in which the training data of the domain is missing. We show that the small amount of in-domain training data that was informed by linguistic hard and soft constraints and domain constraints achieved a label accuracy of 90% in the domain data, thus outperforming state-of-the-art parsers.

Supertagging for domain adaptation: an approach with law texts

The article discusses the problem of automatic detection of amendments found in the Polish statutory law. We treat the problem as a token-classification task and we introduce a scheme constructed by analysis of more than 200 amending bills. We apply recent neural architectures such as BERT and BiRNN to the task of token classification. The achieved results of all models are very high as micro average F1 score ranges from 96.3% to 98.2% for BiRNN. The presented solution is a first step towards fully automatic structuring and application of amendments in the Polish statutory law.

Automatic extraction of amendments from polish statutory law

Sentence Boundary Detection (SBD) is one of the foundational building blocks of Natural Language Processing (NLP), with incorrectly split sentences heavily influencing the output quality of downstream tasks. It is a challenging task for algorithms, especially in the legal domain, considering the complex and different sentence structures used. In this work, we curated a diverse multilingual legal dataset consisting of over 130'000 annotated sentences in 6 languages. Our experimental results indicate that the performance of existing SBD models is subpar on multilingual legal data. We trained and tested monolingual and multilingual models based on CRF, BiLSTM-CRF, and transformers, demonstrating state-of-the-art performance. We also show that our multilingual models outperform all baselines in the zero-shot setting on a Portuguese test set. To encourage further research and development by the community, we have made our dataset, models, and code publicly available.

MultiLegalSBD: A Multilingual Legal Sentence Boundary Detection Dataset

We develop computational models to analyze court statements in order to assess judicial attitudes toward victims of sexual violence in the Israeli court system. The study examines the resonance of rape myths" in the criminal justice system's response to sex crimes, in particular in judicial assessment of victim's credibility. We begin by formulating an ontology for evaluating judicial attitudes toward victim's credibility, with eight ordinal labels and binary categorizations. Second, we curate a manually annotated dataset for judicial assessments of victim's credibility in the Hebrew language, as well as a model that can extract credibility labels from court cases. The dataset consists of 855 verdict decision documents in sexual assault cases from 1990-2021, annotated with the help of legal experts and trained law students. The model uses a combined approach of syntactic and latent structures to find sentences that convey the judge's attitude towards the victim and classify them according to the credibility label set. Our ontology, data, and models will be made available upon request, in the hope they spur future progress in this judicial important task."

The Perfect Victim: Computational Analysis of Judicial Attitudes towards Victims of Sexual Violence

In this paper, we focus on automatic extraction of statute facets from legal statutes such as Act documents. We define statute facets to be key specific aspects of a statute which can potentially be used in legal arguments. For example, Section 25F of the Industrial Disputes Act (India) contains statute facets such as workman, employer, retrenchment of workmen, continuous service for not less than one year, etc. Such statute facets are often used by lawyers as part of their argumentation and also by judges for deciding on a case. In this paper, we propose a weakly supervised technique for extracting such statute facets from legal text. We use dependency tree structure to extract candidate statute facets and use BM25 ranking function to determine statute-specificity of these candidates. We propose a set of facet types which enable us to realize the definition of statute facets in a more computational way. We use recent deep learning models in a few-shot setting to predict an appropriate facet type for each candidate. Only those candidates with high statute-specificity and for which a facet type is predicted with high confidence, are selected as acceptable statute facets. We evaluate the extracted statute facets through both direct and indirect evaluation as well as conduct a user-study to get validation and feedback from lawyers.

Extraction and Classification of Statute Facets using Few-shot Learning

Lexical Normalization

Tokenizations

Dependency Parsing

Part-of-speech tagging

Text extraction is the process of pulling specific information or data from unstructured text. As machines need the information to be in a structured form, for example in a table, text extraction serves to identify relevant information in unstructured text and transforms it into a type of structured information that is easier to process. The goal of text extraction or information extraction (IE) is to make the text’s semantic structure explicit so that we can make use of it. More precisely, it is the process of analyzing text and identifying mentions of semantically defined entities and relationships within it. The entities and relationships that are of interest can be detected and extracted with NLP techniques involving named entity recognition, relation extraction, or event extraction.

Text Extraction

The importance of legal precedents in ensuring consistent jurisprudence is undisputed. Particularly in jurisdictions following the Common law, but even in Civil law systems, uniformity in case law requires adherence to precedents. However, with the growing volume of cases, manual identification becomes a bottleneck, prompting the need for automation. Leveraging the capabilities of natural language processing (NLP) and machine learning (ML), our study delves into the potential of automation in identifying similar cases indicative of precedents. Drawing from a unique, substantial dataset of legal cases from an administrative court in Brazil, we extensively evaluated over one hundred combinations of document representations and text vectorizations. Contrary to earlier studies that relied on minimal validation samples, ours employed a statistically significant sample vetted by legal experts. Our findings reveal that models focusing on granular text representations perform optimally, especially when extracting concepts and relations. Notably, while intricate models may not always guarantee superior outcomes, the importance of refining textual features cannot be understated. These findings pave the way for creating efficient decision support systems in judicial contexts and set a direction for future research aiming to integrate technology in legal decision-making. © 2024 The Authors

Textual similarity for legal precedents discovery: Assessing the performance of machine learning techniques in an administrative court

In judicial practice, large-scale judicial data, transcript information and judicial documents provide empirical guidance for studying historical cases, analyzing criminal situations and adjudicating case results. How to use these data efficiently and accurately to help legal professionals in the judicial field read and analyze texts has become a key issue that needs to be solved at present. This paper adopts the BERT model to obtain dynamic representation word vectors. At present, there are few studies on the combination of BERT and intelligent legal affairs. For traditional pre-training models, its main feature is that it can solve the problem of polysemy in different language environments, and can also obtain the features between long-distance words, so as to better understand the original meaning of complex texts. The BERT-BiLSTM-CRF model based on dynamic representation word vector combination adopted in this paper has achieved a high accuracy rate, and the final harmonic mean FB1 has also increased to an average of 94.95%. © 2024 Copyright held by the owner/author(s).

Research on Event Entity Extraction in Intelligent Economic Legal System Based on Machine Learning

New technologies for assisted negotiation and conflict resolution in law refer to the use of digital platforms and online tools to facilitate and improve communication and collaboration between parties in conflict. These technologies include, among others, online mediation, virtual negotiation rooms, the exchange of information and documents through secure platforms, and the use of Artificial Intelligence to analyze data and generate settlement options. It is important to note that while these technologies can offer significant benefits, they also pose significant challenges. Data privacy and security, the fairness of algorithms, and the potential lack of access to these technologies by certain population groups are critical issues that need to be carefully and thoughtfully addressed. The objective of this paper is to provide a clear and balanced view of the opportunities and challenges associated with new technologies in legal dispute resolution. With this, it is hoped to contribute to the development of effective and ethical strategies for their implementation in the field of law, thus promoting a more accessible justice. In this way, the work is divided into two parts: in the first part, the theoretical framework of online negotiation and dispute resolution (ODR) is reviewed, and in the second part, through the development of three prototypes, we will try to demonstrate a practical solution where we will use technology to serve people and their judicial consultations. © 2024 Ibero-American Law Institute. All rights reserved.

THE CASE OF ASSISTED NEGOTIATION IN THE FIELD OF CONSUMER LAW FOR BETTER ACCESS TO JUSTICE IN CHILE: LESSONS LEARNED IN THE DEVELOPMENT OF SOFTWARE WITH ARTIFICIAL INTELLIGENCE TECHNOLOGIES; [EL CASO DE LA NEGOCIACIÓN ASISTIDA EN EL ÁMBITO DEL DERECHO DE CONSUMO PARA UN MEJOR ACCESO A LA JUSTICIA EN CHILE: LECCIONES APRENDIDAS EN EL DESARROLLO DE SOFTWARE CON TECNOLOGÍAS DE INTELIGENCIA ARTIFICIAL]

The digitization of public services in Germany is always based on a legal basis (e.g., laws). In the digitization process, first relevant entities in law documents (e.g., actors) are detected, then a list of possible process steps of their interactions is derived. The final process is constructed and transformed to a digital service for citizens and companies. Today, the discovery of custom entities in German law documents is still manual high effort work. In our study, we investigate the capabilities of Large Language Models (LLMs) to automate this task, choose five LLMs from 61 evaluated candidates, and perform prompt engineering to create five different prompt variants with differing parts. We examine the automatic annotation by two LLMs (LeoLM and BLOOM CLP German) in detail and find that the inclusion of more information in the prompts as well as an increased number of examples per prompt are beneficial. We report micro F1-scores for the optimal scenario of 0.91 for BLOOM CLP German, and 0.82 for LeoLM, with a higher balanced accuracy for LeoLM. The results indicate that LLMs have a good potential to perform named entity recognition, especially for supporting legal norm analysis in the context of the digitization of public administration. © 2024 Copyright held by the owner/author(s).

Extracting Legal Norm Analysis Categories from German Law Texts with Large Language Models

This paper aims to automate the process of legislative text analysis from judicial case reports to identify applicable law sections for the cases. The proposed methodology combines supervised Machine Learning and Natural Language Processing and is particularly useful for legislative text analysis. The idea is demonstrated with case studies of the Indian Income Tax Act of 1961. We introduced a dataset of case descriptions with applicable law sections from the case details available at the ‘Legal Crystal’ repository. Since multiple law sections could be applied to one particular case, which makes it a multi-label classification problem, we employ two problem transformer techniques, namely one versus rest wrapper and classifier chain, over the Machine Learning models like Logistic Regression, Naive Bayes, Decision Tree, and Support Vector Machine. Different word embedding techniques like n-gram and tf-idf are combined with classifier models to train and test the dataset. We propose a methodology that includes a data pre-processing block, a pipeline consisting of word embedding, a multi-label wrapper, Machine Learning Models, and an evaluation block. Different combination setups of the pipeline are tested for comparative analysis. A customized deep learning model with word2vec and BERT embedding is also tested for the task. It has been observed that the Support Vector Machine with tf-idf word embedding and classifier chain wrapping outperformed the other conventional models and performed at par with the BERT and deep learning model. Although this work is limited to cases related to income tax laws, the proposed methodology is generic to cover any other legal sections. © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2024.

Legislative Text Analysis from Judicial Case Reports Using Machine Learning

In response to the lack of relevant high-quality Q&A (question and answer) datasets in some specific domain Q&A systems, this paper proposes a large language model Q&A Method based on specific domain regulation documents. The paper adopts natural language processing technology combined with a large language model. Firstly, data processing is performed on regulatory documents to extract text that meets the requirements. Then extracting the information of interest from the text and structuring it, and performing vector transformation to store it in a vector database for building a knowledge base. At the same time, performing vector transformation on user's question, and using cosine similarity to calculate the distance between the problem vector and the vector in the knowledge base. Finally, a random method is proposed to form the problem context, and the Lang Chain tool is used to return the Q&A results, which are then displayed in the front-end. Through system design and engineering application, this paper proposes a large language model Q&A method based on specific domain regulation documents, which can accurately obtain Q&A results. © 2024 ACM.

Research on Large Language Model Q&A Method Based on Specific Domain Regulation Documents

This paper studied a legal text extraction algorithm based on segmentation. Considering that manual pixel-level text annotation is usually very expensive and time-consuming, and the existing data sets contain a large amount of text box annotation information, this paper proposed to use text box annotation as auxiliary data for training. To achieve this goal, a dual-task mutual guidance network is proposed, which shares an encoder and two decoders to output pixel-level text segmentation and text box segmentation, respectively. The output of the two decoders can be used as the guidance information of each other, and the output of the pixel-level text segmentation decoder can be used as the guidance information of the text box segmentation decoder to improve the performance of text box segmentation, and vice versa. Experiments on standard data sets show that the mutual guidance network proposed by the algorithm can effectively extract text information. At the same time, the obtained pixel segmentation results can be used to further improve the text recognition performance. © 2024 ACM.

Research on Legal Text Extraction and Recognition Based on Deep Learning

Entity recognition and extraction from contracts play a crucial role in automating contract analysis and extracting valuable information. Named Entity Recognition (NER) techniques are used for identifying and classifying specific entities such as parties, dates, amounts, and clauses within contracts. In this study, we create a high-quality NER dataset from various types of English language contracts by considering their structure, and the legal terminology used within these documents. We present a systematic approach to manually annotate contracts with appropriate entity labels, ensuring accuracy and consistency. The resulting NER dataset serves as a valuable resource for training and evaluating NER models for contract analysis tasks. We evaluate the performance of NER on this dataset using a range of methods. These methods include Conditional Random Fields, various Bidirectional LSTM configurations, and BERT models. Each of these models brings different strengths and capabilities to the task of entity recognition, allowing for a comprehensive evaluation and the selection of the best models over the dataset. Among these, the NER model based on Contracts–BERT–base from the Legal–BERT family, which is pre-trained specifically on English contracts, outperformed all others, achieving an impressive overall F1 score of 0.94. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2024.