Casetext
Litigation: draftingSearch: case lawSearch: legislation
casetext.com
Contents
- What does it claim to do?
- Substantiation of claims & potential issues
- How might the end-user assess effectiveness?
- What form does it take?
- Is it currently in use?
- The creators
- Jurisdiction
- License
What does it claim to do?
Casetext offers legal research and drafting tools. It has three products: (1) Casetext Research, (2) Parallel Search and (3) Compose. Parallel Search can be bundled with Casetext Research, is included as part of Compose, and can be purchased as a stand-alone product.
(1) Casetext Research
Claimed essential features
- Database of caselaw for all 50 US states, including federal case law, analyses, statutes, regulations, and administrative decisions.
- Search using keywords (with optional Boolean connectors).
- CARA A.I. finds sources relevant to an uploaded document (e.g. a brief or claim).
- Provides key metadata and excerpts in the display of search results (e.g. ‘bad law’, ‘good law’, concise summaries).
”[CARA A.I. will] rank the list of citing cases according to the factual and legal context of your document.” (Cite-checking the Smart Way: An Interview about SmartCite with Casetext Co-Founder and Chief Product Officer, Pablo Arredondo; archived)
SmartCite™ provides a range of functionality including flagging authorities which have been “reversed, overruled, vacated, withdrawn or superseded in whole or in part”, providing concise summaries of cited cases, highlighting key passages in judgments, showing which language in a judgment has been emphasised in later cases and allowing users to find cases with similar issues. (Casetext SmartCite™; archived)
Claimed rationale and benefits
- Reduce cost and increase efficiency.
- Provide more relevant results.
- Allow flexible approaches to searching.
“Everything you need for legal research at a fraction of the cost … Comprehensive legal database coverage … Search the Way You Like … Scan Search Results Easily … Automate Rote Work” (Casetext is a true legal research replacement for Lexis and Westlaw for only $65/month; archived)
“Make your practice more effective and efficient with Casetext’s legal research suite.” (Easily understand the cases and statutes; archived)
Claimed design choices
- Searches a legal citation graph using a combination of input text and citation patterns.
- The citator filtering algorithm uses techniques similar to BERT and GPT-2.
- CARA A.I. is trained on “several thousand” overruling cases.
CARA A.I. “uses both the linguistic content and citation patterns in the brief to execute a specialized graph search algorithm over Casetext’s legal citation graph.” (Freeing the Law at Casetext; archived)
Casetext uses “techniques similar to BERT and GPT-2 to … improve their citator filtering algorithm.” (How Casetext Uses Artificial Intelligence; archived)
CARA A.I. “was trained on several thousand examples of overruling cases”. (Use of AI at Casetext; archived)
(2) Parallel Search
Claimed essential features
- Search using natural language queries.
- Find results with the same meaning but expressed with different language.
It “changes the game by taking a natural language sentence from the user and returning passages from the law that mean the same thing even when they have no words in common.” (The Machine Learning Technology Behind Parallel Search; archived)
Claimed rationale and benefits
- To allow lawyers to search in a more familiar way than with traditional search techniques.
Parallel Search “applies the most cutting-edge :natural language processing models, trained on the law, and tailored to the way lawyers search, to help you spend less time searching and find cases traditional search technology would miss.” (Compose Home; archived).
Claimed design choices
- Uses artificial neural network :Transformers to search by concept rather than keyword.
“Parallel Search uses advanced :machine learning techniques to extract concepts from sentences, and find matches based on concepts, rather than keywords.” (The Machine Learning Technology Behind Parallel Search; archived)
“The fundamental building block of Parallel Search is a type of artificial neural network called the :Transformer.” (The Machine Learning Technology Behind Parallel Search; archived)
(3) Compose
Claimed essential features
- Compile draft briefs/motions with up-to-date authority.
- Drafts are tailored to specific positions (movant/nonmovant), courts, argument, and legal standards.
- Use Parallel Search to find authority for specific parts of drafts.
“You can draft a brief in support of or in opposition to a motion in just a few simple steps using Compose… click on the motion that you want to make or oppose from the new brief automation library”
“The list of arguments provides topic sentences for different arguments to make in support of your motion… Click the “Add” button to add that argument to your motion”
“To find cases to support a sentence in your brief, use your cursor to select the sentence that you want to use as your search query.”
(Can you provide step-by-step instructions for getting started on a Compose motion?)
Claimed rationale and benefits
- To provide an “automated associate” for drafting briefs.
- To reduce time spent writing briefs and increase legal accuracy.
- To improve the quality of legal authorities used in briefs.
Compose is “like an automated associate assigned to write the first draft of your brief.” (What is Compose?; archived)
Compose cures “blank page” syndrome and starts attorneys off right. (Craft exceptional briefs, without the busy work; archived)
Compose “helps attorneys draft high quality briefs in less time” Compose video (Compose; archived)
Expert attorney editors compile and regularly update the arguments and legal standards on Compose, so you always have the most current and relevant information at your fingertips as you draft. (Craft exceptional briefs, without the busy work; archived)
Claimed design choices
- Compose points to its “[g]roundbreaking :machine learning technology [Parallel Search] paired with the insights of expert litigators, for a streamlined brief-drafting workflow.” (Compose; archived)
Substantiation of claims & potential issues
- It is difficult to assess the performance of search systems. Lawyers may not appreciate that the results returned will vary according to the design of the system, and that it may fail to return relevant results. As a result, they may place too much reliance on the results of a single search system.
- Casetext’s marketing states that it finds cases that Lexis Nexis and Westlaw miss. This should alert lawyers to the fact that legal research systems are not perfect.
- Case summaries for individual cases are constructed from those parts of the opinions of judges in later cases that briefly summarise the earlier case. Such judge-made summaries might only relate to one of several aspects decided in the case.
- Highlighting within results may flag important aspects of cases, but it may also encourage lawyers not to review a case in its entirety.
- Systems like Casetext’s Compose that generate first drafts of legal documents may shape drafting practice, and encourage reliance on the system.
Casetext provides information about many aspects of its technology but there is no single text explaining the workings of all the elements of all its products or providing a detailed exposition of the workflows from collection of the data to output of results. In this section we provide information about various core elements of each of its products and modules.
Casetext Research
Data
- The dataset consists on case law of 50 states, it is not specified how large the dataset is, and whether it includes all available case law.
- There is no technical description available on the website about the original sources (i.e. is data provided directly by the courts or via a third data provider) and the data extraction methods used to add documents to the database.
Search
- As Casetext Research supports boolean and keyword search, at least a part of the system consists of a classic retrieval engine (such as :Elasticsearch).
- Casetext Research supports filtering on metadata such as jurisdiction, motion type, party type, cause of action, and date range. (Cite-checking the Smart Way: An Interview about SmartCite with Casetext Co-Founder and Chief Product Officer, Pablo Arredondo; archived)
- An important technological innovation in Casetext Research is the inclusion of citation relations from the SmartCite™ citator in the retrieval engine.
SmartCite™
- SmartCite™ is Casetext’s citator system. It assesses whether a case returned by a search is still good law. In order to make this assessment ”[Casetext has] our AI read the entirety of every case, and then pass[es] along to our lawyers only the portions of a case that have a chance of overruling another case.”, i.e. it is a semi-automatic process. For the :machine learning part Casetext uses “techniques similar to BERT and GPT-2 to … improve our citator filtering algorithm.” (How Casetext Uses Artificial Intelligence; archived), which amounts to the use of embeddings as document representations. The SmartCite citator employs four different relations: (1) overruled or reversed on appeal; (2) affirmed on appeal; (3) cited as contrary signal; (4) relies on negatively-treated case. (Cite-checking the Smart Way: An Interview about SmartCite with Casetext Co-Founder and Chief Product Officer, Pablo Arredondo; archived)
- For its training data SmartCite™ uses information on “how a case has been cited by other cases.” It relies on various data sources including case summaries offered by judges (“explanatory parentheticals”), phrases used in judgments (such as “overruled by”) and other “signals” in judicial opinions (use of the term “contra” or similar). These data sources are harvested to “extract … case summaries and explanations” from the text of judicial opinions which are then used to train the citator system. (Cite-checking the Smart Way: An Interview about SmartCite with Casetext Co-Founder and Chief Product Officer, Pablo Arredondo; archived) The AI “was trained on several thousand examples of overruling cases” (Use of AI at Casetext; archived) gathered and annotated by the company.
CARA A.I.
CARA A.I. is a module in of Casetext Research and Compose.
Data
- CARA A.I. is both a “document analyzer” and “case law recommendation engine” which ingests briefs or other documents uploaded by users to return results tailored to the documents. Walker states that CARA AI “uses both the linguistic content and citation patterns in the brief to execute a specialized graph search algorithm over Casetext’s legal citation graph.” (Freeing the Law at Casetext; archived).
System
- The modules behind CARA A.I. are not explicitly mentioned on the website or blog posts and articles but the “similar issues” feature is based on the use of embeddings to find similar passages in the database: “…we use a technique for turning a passage of legal text into a high-dimensional vector … For a given passage of text, … our system returns other passages from across the law whose vectors have the smallest angles..” (How Casetext Uses Artificial Intelligence; archived)
- The techniques used to recognise and resolve citations and other entities (names, dates, …) are not explicitly mentioned, but include identifying candidates against the documents in the database:
“Part of our entity resolution processing involves making many millions of computationally expensive, full-text search queries. Originally, we did these queries against a very large :Elasticsearch cluster, but we recently swapped out :Elasticsearch for raw Lucene indexes. This gave us a well over 10x improvement in processing throughput for our batch pipelines.” (Freeing the Law at Casetext; archived)
Parallel Search
“Parallel Search uses advanced :machine learning techniques to extract concepts from sentences, and find matches based on concepts, rather than keywords… The fundamental building block of Parallel Search is a type of artificial neural network called the :Transformer.” (The Machine Learning Technology Behind Parallel Search; archived)
- As in CARA A.I. the retrieval process is based on similarity between the document representations (embeddings) that are present in the database and the transformed queries.
- Casetext uses traditional information retrieval measures to measure the performance of the Casetext Research and Parallel Search engines, i.e. Precision, Recall, and non-discounted cumulative gain (nDCG). (Casetext, Search Results Evaluation Efforts at Casetext; archived)
Compose
- Compose is a research and drafting tool. It is a document generation system that uses templating from a library of motion templates curated and updated by an in-house team of lawyers, which is fully accessible in the tool (tested in a demo period). It also includes access to the Parallel Search feature to search for (conceptually) similar phrases or texts in the database.
Resources
- P.D. Arredondo, “Harvesting and Utilizing Explanatory Parentheticals” (2015) 1 Legal Information Review 31 (relevant to the use of “explanatory parentheticals” by Casetext’s SmartCite™ citator system).
What form does it take?
Form
Application, Platform
Details
Casetext describes two of its products, Casetext Research and Compose, as platforms. It describes Parallel Search as an application. (Casetext: About; archived) Parallel Search is included as part of the Compose Product, can be bundled with the Casetext Research product or purchased as a stand-alone product.
Casetext offers customised enterprise solutions for firms “of 11 or more”. (Casetext; pricing; archived
Top Is it in current use?
Casetext states that it is “[t]rusted by 8,500 firms, from solo practitioners to Am Law 100 firms.” (Casetext: Home; archived)
Top The creators
Created by
Legal tech company
Details
Casetext Inc., is based in San Francisco, California. It was founded in 2013. Its co-founders include Jake Heller, its CEO, and Pablo Arrodondo, its Chief Operating Officer. Both were litigators. (Casetext: About; archived)
Top Jurisdiction
Background of developers
US
Target jurisdiction
California, US
Target legal domains
Casetext provides access to all state and federal cases, statutes, regulations, and rules. It also provides access to the administrative decisions of the following:
Tax Court, Board of Immigration Appeals, Board of Tax Appeals, Equal Employment Opportunity Commission,, Patent Trial and Appeal Board, Trademark Trial and Appeal Board, National Labor Relations Board
Full details of coverage can be found at Casetext legal database coverage (archived).
Top License
Casetext has github repos but it’s not clear how much of the code in the repos was developed by Casetext. The github license provides that “If you set your pages and repositories to be viewed publicly, you grant each User of GitHub a nonexclusive, worldwide license to use, display, and perform Your Content through the GitHub Service and to reproduce Your Content solely on GitHub as permitted through GitHub’s functionality (for example, through forking).” Subscribers can grant broader rights to users by adopting an open source licence. The code in several of the Casetext repos is subject to an open source license (MIT/Apache 2).
Top