Generative AI and multilingual language processing: Changing the game for rule-makers and rule-takers
23 Jul 2024
6 mins
The latest advancements in artificial intelligence (AI) are revolutionizing multiple domains, from language processing to the creative arts. Today’s AI tools can craft original artworks, enhance image quality, restore damaged photographs, and compose music in diverse styles and genres.
A pivotal breakthrough lies in the development of Large Language Models (LLMs) and generative AI which excel at summarizing large texts – a fundamental advancement for regulated industries where significant effort is devoted to transforming lengthy documents into concise, operational forms.
Summarization: A mainstay for rule-makers and rule-takers
While we often talk about publishing legislation, regulations, standards, and official guidance in linear forms, suitable for reading from start to finish. In reality, non-linear readings are the norm.
Take the United States’ Internal Revenue Code (IRC) which underpins the Internal Revenue Service (IRS) in administering and enforcing tax laws. IRS staff have created their own distilled guidebook of that very large and complex corpus, known as the Internal Revenue Manual (IRM). The IRC is literally the letter of the law but the IRM is much more useful for practitioners as it summarizes, explains and guides the interpretation of the IRC in a way that is much more attuned to how the tax SMEs downstream actually think when doing their work.
Downstream of the IRM, tax attorneys, CFOs, etc. create internal policies, memos, and procedures to reflect the latest updates in the IRM and thus the IRC.
In engineering, professionals distill extensive amounts of standards to identify key requirements for compliance. Legislative and parliamentary officials depend on legislative and fiscal analysts to condense legislation into actionable summaries to support their decision-making. These summarizing documents have names such as explanatory memoranda, fiscal notes, bill summaries. Much of this summarized information is made publicly accessible.
Summarization activities – production and consumption – make up a substantial aspect of the daily work for subject matter experts (SMEs) on both sides of the rule-maker/rule-taker coin. Thus, the advanced summarization capabilities of LLMs present significant opportunities to rule-makers and rule-takers alike.
Tackling writer’s block
Professionals tasked with writing summaries understand the daunting challenge of starting from scratch, the blinking cursor and the blank page.
The key determinant of the quality of automated summarization technology’s output is whether it provides a solid first draft for the SME to refine – a good step in the right direction that does not need to be reworked so much that its value is questionable. Before LLMs, the jury was out on this. Now, however, it is absolutely clear that the quality of the summarization output is typically more than good enough for a first draft and sometimes excellent.
Multilingual translation and simplified language use cases
We are seeing an increasing interest in using linguistic machine translation technology in summarization use cases, for example, taking 150 pages of English text and summarizing it into five pages of English.
The term ‘explain like I’m five’ is becoming synonymous with the use of AI algorithms to achieve this.
In addition, in heavily regulated industries we are seeing increasing interest in leveraging AI to generate Simplified Technical English (STE) and other controlled languages. Examples range from train drivers to legislative drafters.
By combining language summarization with language translation, a plethora of useful features emerge. For example, the ability to take an 800-page document in Japanese and create a summary of that document in English. Use cases for this capability range from patent attorneys to medical writers.
Managing regulatory change with AI
Another area in which LLMs and summarization can solve pain points in regulated industries is by helping to address the regulatory change management challenge.
On a day-to-day basis, knowledge management professionals face the challenge of pinpointing impactful changes within updates to extensive legislation/regulation/standards/guidance. Editorial changes may affect the text superficially without changing the meaning. An example would be when text is re-arranged. When this happens, there is a high probability that the machine-generated summaries of the original and updated texts will be substantially the same.
Differences in machine-generated summaries can thus highlight so-called ‘material’ changes, helping professionals discern whether updates necessitate action, such as revising a team training package.
This automation of materiality detection is key and comes up repeatedly as a pain point. SMEs may already have the ability to see a timeline of changes but now with summarizations, this timeline can be annotated to indicate materiality and thus greatly expedite the endless cycle of regulatory change management.
Automating tense shifting for regulatory filings
Companies operating in regulated industries often need to send submissions to regulators to get approval ahead of actions such as carrying out a clinical trial for a drug, getting planning permission for construction of an airport, etc. The submissions are phrased in the future tense. However, post-completion, the company must typically write extensive reports and file these with regulators.
LLMs’ ability to take future tense language and transform it into the past tense can thus save substantial time and effort.
Unlock the future of regulatory compliance
AI undoubtedly has the potential to help professionals in regulated industries streamline workflows, manage regulatory changes, and simplify compliance processes. At Propylon, we have the specialist skills and expertise to position your organization for the future. Learn more about what we do.
Sean McGrath
All from Sean McGrathRelated Articles
