Speech Act and Pragmatic Analysis

Overview

Speech act analysis classifies texts by their communicative function -- what the author is DOING with language (asserting, advising, explaining, questioning, challenging, agreeing/supporting) rather than what they are saying. By measuring the proportion of each speech act type across a corpus, the analysis produces a pragmatic signature: a distributional profile of how an author uses language to act in the world. The core principle: speech acts reveal communicative habits that persist across topics -- a person who predominantly explains and qualifies does so whether discussing technology, politics, or cooking. These proportions produce replicable constraints for voice matching that complement lexical, sentiment, and structural analyses.

Research foundation: Speech act theory originates with Austin (1962, How to Do Things with Words) who distinguished locutionary (saying), illocutionary (doing by saying), and perlocutionary (effect of saying) acts. Searle (1976) systematized this into five illocutionary categories: assertives (committing to truth), directives (getting hearer to do something), commissives (committing speaker to action), expressives (expressing psychological states), and declarations (bringing about states of affairs). The CMC Act Taxonomy (Herring, Das, & Penumarthy, 2005) adapted speech act theory to computer-mediated communication with 18 act types suitable for online discourse. For pragmatic analysis of written corpora, a hybrid approach combining dictionary-based signal detection with contextual classification produces the most reliable results (Zhang et al., 2011; Qadir & Riloff, 2011).

When to Use

• Categorizing posts, comments, or documents by what the author is doing (asserting, explaining, questioning, etc.)
• Measuring proportions of speech act types across a corpus to build a communicative profile
• Comparing how an author's communicative function shifts across topics, time periods, or contexts
• Building pragmatic constraints for voice replication ("this author advises 25% of the time, questions 15%")
• Identifying whether an author predominantly informs, persuades, supports, or challenges
• Complementing lexical and sentiment analyses with a functional layer

When NOT to use: