311 receives long resident complaint emails and must turn them into structured topic labels for routing. Which text-analysis capability should they apply?

Summarization combined with topic detection converts long free-text complaints into structured topic labels that routing systems can consume. 311 receiving lengthy resident complaint emails should summarize the content and detect topics so each message becomes labeled routing signals—trash, noise, road hazard, and similar municipal categories—rather than left in whatever form a different kind of analysis produces. That text-analysis pairing is how unstructured email becomes assignable work.

Translating the emails and detecting language to produce structured topic labels identifies the writing language and produces translated prose, not the topic labels a router needs to pick a desk—a perfectly translated complaint about a pothole still has to be topic-tagged before it can be routed. Scoring sentiment and mining opinions in emails to produce structured topic labels tells 311 how angry or happy residents sound and which aspect they reacted to, which is a useful attitude signal, but a sentiment score with mined opinions is not a topic label and cannot tell dispatch whether the issue is trash, noise, or a road hazard. Detecting and redacting PII in the emails to produce structured topic labels strips names and contacts for privacy, which is good data hygiene, but stripping PII does not generate the topic labels routing depends on—a redacted email is still unlabeled.

Exam caveat: topic models and LLM labels can oversimplify multi-issue emails; allow multi-label output and human re-route when residents describe several problems. Summaries must not drop safety-critical details. Operational check: summarize and topic-label a gold set of complaint emails, compare labels to supervisor routing decisions, enable multi-topic tags where needed, and monitor misroute rates after deployment.