Endofutterance

Endofutterance is a boundary marker in discourse that signals the conclusion of an utterance by a speaker. In linguistics and computational linguistics, an utterance is the spoken or written unit that carries a complete thought or discourse function within a turn in conversation. The endofutterance boundary helps segment continuous speech into analyzable units and supports dialogue management, transcription, and language processing tasks. It is often discussed in the context of segmentation in spoken corpora and in the design of interactive systems.

In spoken language, endofutterance is realized through prosodic cues such as pause duration, an intonation fall, perceived finality, and speaker breath. In written text, punctuation marks like periods, question marks, and exclamation points often convey an endofutterance. However, in practice boundaries may not align perfectly with sentence ends; utterances can span multiple sentences or consist of incomplete phrases.

In NLP and speech technologies, detecting endofutterance is essential for streaming automatic speech recognition (ASR) segmentation, dialogue systems, and corpus annotation. Approaches include rule-based cues, prosodic feature analysis, and supervised learning with models such as conditional random fields or neural networks trained to label endofutterance boundaries. Some datasets annotate utterance boundaries explicitly with an endofutterance tag, while others rely on punctuation or speaker turns.

Challenges include overlapping speech, interruptions, backchannels, rapid turn-taking, and noisy environments that complicate boundary detection. Cross-linguistic variation in prosody, speaker variability, and conversational style affect boundary timing. Evaluation often uses boundary placement error rate or comparisons to hand-annotated gold standards.

See also: end of sentence, utterance boundary, boundary detection.