How Generative AI Can Address Language and Data Challenges in Complex Litigation

In a recent U.K. High Court litigation involving parties across multiple jurisdictions, counsel faced the challenge of reviewing a large volume of traditional electronic documents and hard copy, including handwritten documents in numerous languages under demanding court deadlines. To respond to discovery obligations, a team of e-discovery experts and data scientists, working closely with outside counsel, established a series of workflows using large language models and generative AI to identify relevant documents, accelerate review, address language barriers and validate results for defensibility. 

When legal teams face compounding challenges — large volumes of potentially relevant documents, tight court-imposed deadlines and underlying data that may include poor quality scans, handwritten materials and content in languages that require translation — it becomes difficult, and in some cases functionally impossible, to conduct a defensible review using traditional methods. Generative artificial intelligence can provide relief in these scenarios, though its effectiveness hinges on how it is applied. When deployed with expert oversight and tailored to the specific demands of a matter, it can accelerate key stages of review. 

Often, the matters that benefit most from an AI-assisted approach are those where the document population is large, complex and/or spans languages and formats. In these situations, a flexible, integrated workflow combining multiple technologies is essential to complete an efficient review within the constraints of the case.

When managing large document populations that contain multiple languages, human translation at scale is time-consuming and expensive, and relying on it as the primary method for surfacing relevant materials is rarely proportionate to the case. With proper direction, large language models are capable of rapidly analyzing, summarizing and translating multilingual content. This can allow legal teams to perform initial relevance assessments across a multilingual dataset without waiting for full translation of every document. AI-assisted workflows can surface documents that warrant closer attention, allowing reviewers to focus their time on materials most likely to be significant to the case.

Because each language is not equally represented in the datasets used to train various large language models, translation quality depends on the languages in the review and the models being used. This is one of many reasons why expert oversight is critical, to help identify gaps, adjust models and fine-tune workflows before important information is overlooked or misinterpreted. 

Legal teams also often encounter documents that cannot be readily processed and loaded into a review platform. Examples include handwritten materials that may hold relevant context, poor quality scans of paper documents or images and drawings that can’t be searched for text. Advanced optical character recognition and AI-powered extraction capabilities can convert these materials into reviewable formats, so they may be analyzed alongside and in the context of the full document set. Additionally, generative AI can further support extraction and analysis of metadata to enhance insights and reduce the volume of data requiring manual review.

For these approaches to work, and to produce reliable, defensible results, the technology must be used in a way which addresses the specific characteristics of the case and the data involved. Equally important is that the workflows be designed and led by experts who know how to tune the models and validate the outputs.  

In the U.K. High Court litigation mentioned earlier, the team applied numerous technologies in parallel to overcome the unique language and document challenges. The methodology included:

• Continuous active learning models to analyze documents identified as potentially relevant, helping to reduce the overall population and provide ongoing quality control throughout the review.
• AI-powered objective coding and key metadata extraction from PDFs, advanced optical character recognition to convert poor quality scans into a reviewable format, and translation capabilities for documents in foreign languages.
• Systematic examination for hallucinations to identify flaws in AI-generated outputs and continuously improve the system for greater accuracy.

This enabled the legal team to conduct review under timelines that would not have been achievable through traditional means. It provided significant cost and time savings across first-level review compared to conventional approaches; and greater accuracy at each phase of review than standard tools would have provided given the data formats and language elements. The combination of expertise with advanced technology helped to reduce the risk of missing relevant materials due to language barriers or document quality issues and provided results that could withstand scrutiny from courts and opposing counsel.

As AI capabilities continue to develop rapidly, organizations engaged in complex, data-intensive matters have an increasing opportunity to leverage these tools effectively. The firms and legal teams that will benefit most are those that embrace creative applications of the technology, customize tools according to each specific case and continue to prioritize expert oversight.

Related topics:

The views expressed herein are those of the author(s) and not necessarily the views of FTI Consulting, its management, its subsidiaries, its affiliates, or its other professionals.