Surface and Contextual Linguistic Cues in Dialog Act Classification: A Cognitive Science View.

Area of Science:

• Cognitive Science
• Computational Linguistics
• Natural Language Processing

Background:

• Dialog act classification is a computational task analyzing utterance intention.
• Existing studies often focus on model performance without detailing cue importance.
• Cognitive science offers a framework to investigate human interpretation of linguistic cues.

Purpose of the Study:

• To investigate the role of surface and contextual linguistic cues in identifying utterance intention.
• To compare the explanatory power of different cue types in dialog act identification.
• To understand how humans and computers utilize linguistic cues for speech act recognition.

Main Methods:

• Review of computational dialog act classification studies and corpora.
• Cognitive science approach using logistic regression models.
• Analysis of explained variance for surface and contextual cues, including word frequency statistics.

Main Results:

• Surface linguistic cues are foundational for human dialog act identification.
• Contextual cues complement surface cues, but are less dominant.
• Word frequency statistics significantly contribute to identifying dialog acts.
• Consistent trends were observed across diverse corpora and dialog types.

Conclusions:

• Surface linguistic cues play a critical role in human speech act recognition.
• Understanding cue importance enhances both human and computational dialog understanding.
• Linguistic cue analysis provides insights into cognitive processes of communication.