What is natural language processing?

Definition

Natural language processing is the branch of AI focused on language: reading it, classifying it, extracting facts from it and writing it. Large language models are the newest and most capable NLP tools, but older NLP still runs plenty of software, like email classification and keyword surveillance.

FINRA's Notice 24-09 lists natural language processing among the technologies the term AI covers.

Why it matters for advisors

Compliance teams have used NLP for years in archiving and surveillance tools that flag risky language in emails and chats. Newer tools use language models to do the same with more nuance and fewer false alarms.

Knowing the term helps when a vendor says its review is "NLP-based": ask whether that means keyword rules, a trained classifier or a language model, because the risks and the testing differ.

Older NLP features, like keyword-based email surveillance, are often being replaced by LLM-based ones. If your archive or review vendor switches, ask how they tested the new approach and whether it catches what the old one did.

Example

A firm's archiving vendor offers an AI review feature for flagged messages. The CCO asks for false-positive and false-negative rates from a test on the firm's own sample before switching it on.

Watch out for

Automated review supports supervision. It doesn't replace the person responsible for it.