Update evaluate protocol

manuscript/main/main.tex

@@ -231,25 +231,31 @@ camera-wearer, and aggregate non-wearer activity. Second, the original
 participant timeline is retained to derive committed interaction events for
 behavioral evaluation.
 
-We consider two wearer-relative events. An \emph{EOT-response} denotes a
+An \emph{EOT-response} is an observed wearer take-turn after a non-wearer floor
-non-wearer floor release followed by the wearer taking the floor, whereas an
+release; an \emph{INT-yield} is an observed wearer floor release after a
-\emph{INT-yield} denotes a non-wearer takeover while the wearer is speaking,
+non-wearer takeover during wearer speech.
-followed by the wearer relinquishing the floor. Candidate events are generated
+
-automatically from participant-level onset/offset annotations. Mid-turn pauses,
+\textbf{Hard-negative region definition.}
-other-to-other handovers, backchannels, and non-takeover overlaps are retained
+Committed positive events and hard-negative regions are derived separately
-as hard-negative regions rather than positive events. Annotation-corrupted or
+from the participant-resolved speech timeline. Positive EOT-response and
-unresolvable boundaries are excluded from scoring. Ambiguous candidates are
+INT-yield events are anchored by observed wearer behavior. In contrast,
-manually adjudicated.
+hard-negative regions do not require any wearer transition. For example, a
+mid-turn pause followed by the same non-wearer resuming speech, or an
+other-to-other floor transfer with the wearer remaining silent throughout,
+defines an EOT hard-negative region: a take-turn commitment within this region
+is counted as a false positive. Likewise, brief backchannels or non-takeover
+overlaps during wearer speech define INT hard-negative regions. Ambiguous or
+annotation-corrupted regions are excluded from scoring.
 
 Participant identity is used only to construct these evaluation annotations
 and is never provided to the prediction model. Provided person boxes are used
 only for the causal visible-person input; transcripts, target-of-speech labels,
 close-microphone audio, and future observations are unavailable to the model.
 \todo{Report duration and event statistics: EOT positives/hard negatives,
-INT positives/hard negatives, excluded candidates, and manual-adjudication
+INT positives/hard negatives, excluded candidates, and audited-region counts
-count for train/development and test. Freeze the event-construction and
+for train/development and test. Freeze the event-construction and region-audit
-adjudication protocol before reporting semantic EOT/INT results; until then,
+protocol before reporting semantic EOT/INT results; until then, retain the
-retain the timing-derived proxy terminology.}
+timing-derived proxy terminology.}
 
 \subsection{Model Configuration and Training}
 
@@ -276,12 +282,14 @@ EgoSpeak, reporting AP at the 0.2--2.0~s horizons and their mean mAP.
 
 Behavioral evaluation follows a committed-event protocol. For both
 EOT-response and INT-yield, each gold event is matched to at most one committed
-prediction and is scored by recall and signed detection latency. A prediction
+prediction and is scored by recall and signed detection latency.
-inside a task-specific hard-negative region is counted as a false positive.
+
-EOT hard negatives include mid-turn continuations and competing
+\textbf{False-positive scoring rule.}
-other-to-other floor transfers; INT hard negatives include backchannels and
+A false positive is defined by the model committing inside a labeled
-overlaps in which the wearer retains the floor. Regions excluded during
+hard-negative region, regardless of whether the wearer subsequently exhibits
-adjudication do not contribute to either the positive or negative denominator.
+an onset or offset. Thus, false-positive evaluation includes conversational
+situations in which the correct wearer behavior is to remain silent or retain
+the floor.
 
 We report the recall--false-positive trade-off and a common operating point at
 matched recall $R=0.8$. The primary behavioral quantities are therefore EOT