Intent recognition

Intent recognition is the language processing step that works out what a person actually wants from what they say or type. A caller who asks "Do you have anything tomorrow afternoon?" wants to book an appointment; one who says "My order still hasn't arrived" wants a status update, or is about to complain. If the intent is misread, the whole conversation heads in the wrong direction. In voice systems, this step comes right after the caller's speech is turned into text.

How it works

The classic approach starts with a predefined list of intents and example phrases for each; a classifier then maps every incoming sentence to one of those intents. Covering a new type of question means adding a new intent and new examples. With a large language model (LLM), the model interprets the sentence in context instead of being tied to a fixed list. Either way, details such as dates, times and names are extracted alongside the intent. When the intent is unclear, a good system asks a clarifying question rather than guessing.

Why it matters

In an IVR menu, callers pick their intent themselves by pressing a key; with a voice agent, the system has to understand it. Whether the agent gives the right information, starts the right task and hands the call to the right team all depends on this step. In Turkish, suffix-heavy words and flexible word order mean intent recognition has to be tested with real sentences.