uttags

Uttags is a term used in linguistics and natural language processing to refer to a system of labels applied to utterances in conversational data. The aim is to annotate the illocutionary or pragmatic function of each utterance, enabling analysis of dialogue structure and the training of dialogue systems.

Tag sets for uttags are typically multi-layered. A coarse-grained layer might classify utterances as statements, questions, commands, or exclamations. A finer-grained layer can capture dialogue acts or pragmatic intents such as information requests, confirmations, directives, greetings, backchannels, or topic-management acts like topic opening or topic closing. Annotation guidelines address edge cases, such as sarcasm, ellipsis, or code-switching, and often emphasize consistency across annotators and contexts.

Usage and impact: Uttags are used in annotated corpora to study conversational dynamics, evaluate dialogue systems, and extract features for machine learning models. They facilitate comparisons across languages and genres and support tasks such as intent recognition, response selection, and discourse analysis. Reliability is typically assessed with inter-annotator agreement measures, and researchers may employ hierarchical or language-specific adaptations of the tag set.

Examples: "What time is it?" can be tagged as UTT-Question; "Please send the file." as UTT-Command/Request; "Thanks." as UTT-Gratitude; "Uh-huh." as UTT-Backchannel; "See you later." as UTT-Closing.

Relation to other concepts: Uttags relate to dialogue act tagging and speech-act theory but focus specifically on utterance-level labeling within dialogue data. They are one of several annotation schemes used to structure and interpret conversational content.