Transcription has a measurable objective: compare the output with the words that were spoken. Intent interpretation has no equivalent ground truth. The same sentence can suggest an observation, a proposal, or an obligation, so extracting intent is a problem of classification under context, not simply accuracy against a reference text.
Cino is designed around that distinction. The system must preserve uncertainty where language does not justify a task, while still identifying commitments that a distributed team can act on.
Why Intent Is Harder Than Transcription
A transcript answers, "What words occurred?" Intent extraction asks, "What should happen because they occurred?" The second question depends on speaker, role, surrounding discussion, and whether the statement describes a decision or a possible future action.
Language rarely labels its own function. "We should look at that" can be a suggestion, a request for investigation, or a polite way to defer the issue. A system that turns every such sentence into a task creates work that nobody accepted. A system that ignores every ambiguous sentence misses useful commitments.
This is why Cino treats meeting understanding as several linked decisions: identify who spoke, interpret the conversational function, detect commitment strength, and retain only an actionable result. Each step narrows the set of statements that should become tasks.
Attribution Starts With the Speaker
A task without an owner is incomplete. Speaker diarization separates voices in the recording, but useful attribution also requires mapping a voice to a participant and preserving who made the commitment.
Attribution is not a cosmetic transcript feature. "I will send the draft" and "Priya will send the draft" have different owners, even when the action is identical. Overlapping speech, interruptions, and similar voices can make the distinction uncertain, so Cino treats speaker identity as input to task extraction rather than an afterthought.
The surrounding setting supplies additional evidence. Role-aware attribution, professional meeting language, and distributed team context are three commitment markers that help distinguish a casual remark from an assigned action. None is sufficient alone, but together they make the interpretation more disciplined.
Recognizing Commitments in Natural Language
Commitments often appear without an imperative. "We should look at that, and I can have it done by Monday" contains both a topic and an explicit owner, action, and time. That is materially different from "We should look at that", which leaves the responsible person unresolved.
The linguistic signal is distributed across the sentence. First-person ownership, an action verb, a concrete object, and a time reference can combine into a task. Cino evaluates these parts together instead of matching isolated words such as "should" or "need", which occur in suggestions as often as they occur in commitments.
The distinction also works in the negative direction. "Shouldn't someone look at that?" raises a possible responsibility but does not assign it. In contrast, "Priya will implement option A by end of sprint" names both the action and its owner. The system should preserve that difference rather than normalize every request into the same task type.
Decisions Are Not Always Commitments
"We are going with option A" records a decision about direction. It does not, by itself, say who will perform the next action or when it will happen.
A decision can be valid meeting output without becoming a task. Treating the two as identical produces task records with no accountable owner.
The next sentence may add the missing commitment: "Priya will implement option A by end of sprint". Here the decision and the task are related, but they are not interchangeable. Cino keeps the distinction so notes can represent what the team chose without inventing work that was never assigned.
What Cino does not do
Cino identifies explicit meeting commitments; it does not act as a general project manager. We are not saying that every important statement should become a task. We are saying that task creation should follow evidence in the conversation.
First, Cino does not infer priority from meeting language. A speaker may call something important without specifying whether it is urgent, high priority, or simply worth recording. Those labels require a separate decision from the team.
Second, Cino does not resolve conflicts between commitments. If two statements assign incompatible actions or dates, the disagreement remains visible for people to settle. Selecting one interpretation would create an unsupported decision.
Third, Cino does not infer deadlines from vague temporal language. "Soon" and "when you get a chance" do not produce deadlines. A system can record the commitment while leaving timing unspecified, rather than turning conversational softness into a date that nobody stated.
Why Precision Matters More Than Recall
For meeting tasks, precision and recall pull in different directions. Higher recall captures more possible commitments but also admits more false positives. Higher precision may leave some ambiguous actions unlisted, but the tasks that appear are more likely to be trusted.
After a forty-minute call, receiving "five real tasks and three false positives per call" creates review work and weakens trust in the task list. People must check whether each item is real before they can act on any of them.
Cino tunes toward precision, so it may miss vague or implicit commitments rather than create spurious tasks. Recall can improve naturally as teams develop clearer commitment language in their meetings.