Neural Nets & Document Search: Game-changing AI for Litigators

Google+ Pinterest LinkedIn Tumblr

How often are you searching large collections of documents or trying to find a specific statement in a document? For most litigators, the answer is every day. Whether digging through discovery, transcripts, or internal brief banks, litigators and their teams are spending time—lots of it—on searching documents. 

Traditional Search Tools & Keyword Prison

Legal professionals rely on search engines to find relevant information. While these tools do help to locate relevant materials, they are often cumbersome, requiring users to manually come up with ideal search terms expressed as keywords or Boolean, in order to locate the records most relevant to the matter at hand. 

Litigators find themselves at the mercy of their ability to identify the perfect keywords to conduct their search. Keyword searches will then return thousands, if not millions, of possible documents to review. Attorneys spend hours sifting through result lists to find cases and other documents relevant to the pending litigation.  

Essentially, litigators find themselves locked in keyword prison. 

AllSearch & The Application of Neural Nets to Natural Language Processing 

Enter AllSearch, the concept-based document search tool developed by Casetext to free these attorneys from lockup. AllSearch is technology that leverages breakthroughs in applying transformer-based neural networks (a form of AI) to natural language processing (NLP, another subset of AI). Let’s break that down a bit.

Neural nets have been around for decades, but there’s been a recent revival of this technology that includes the application of neural nets to NLP. Within the neural net family, there are transformer-based neural nets. In 2018, Google made a fundamental breakthrough when it developed BERT, a transformer-based machine learning technique for NLP pre-training that applied neural net techniques to capture language in a whole new way. 

Casetext immediately set about applying this technology to build AllSearch, which uses neural nets to read any document. AllSeach users upload the document or body of documents they wish to search. Instead of storing documents as a “flat” keyword search index, AllSearch stores documents and their language into a vast, dimensional vector space. In this space, AllSearch uses neural networks to go beyond merely identifying keywords to match actual concepts in documents—even if the words have no overlap—because it 

Speed, Accuracy & Versatility for Litigators 

AllSearch allows litigators to apply the power of these technological advances to search anything with greater accuracy and speed. With AllSearch, attorneys don’t need to spend time thinking about ideal keywords or construct perfect queries for searches. Instead, they can simply input a sentence or phrase summing up the concept they are looking for, which saves them time. AllSearch will then match attorneys’ queries by concept rather than by literal keyword.

Concept-based search is more accurate and produces less irrelevant results. This means less documents for litigators to dig through to find the relevant document or information, so they can find what they need faster and cuts down on review time.

AllSearch is a versatile search tool that can be applied to save litigators significant amounts of time when searching documents. Law firms can use AllSearch to quickly search internal document repositories such as brief banks, thereby streamlining knowledge management. AllSearch can also be used to expedite e-discovery, efficiently search transcripts and litigation records, and more. 

To learn more about AllSearch and its capabilities, visit

Casetext automates litigation tasks to help attorneys efficiently provide high-quality representation.