Pragmatic Language Interpretation as a Probabilistic Inference
Grice’s theory of implicature is comprised of the cooperative principle and several rules of conduct. The maxims are as follows: quantity, quality, relation, and manner. The cooperative principle states that one can assume that in an ordinary conversation, the other discourse participants are being cooperative, i.e. brief, clear, orderly, and avoiding ambiguity. One of the more important quandaries in understanding language is understanding a conversational implicature rather than an entailment. An entailment is not a cancellable statement, it is fact and provides the foreground of a claim made by a speaker. An implicature is a cancellable statement and is inferred from Grice’s cooperative principle. The at-issue here is building a model to understand the “educated guess assumption” being made by the listener in light of an implicature based on the speaker’s words, tone, body language, facial expression, etc. Goodman and Frank describe the “probabilistic approach to pragmatics;” understanding conversational utterances through a rational speech act model. The model considers inference in addition to the explicit word choice and sentence structure most commonly considered in semantics. In one of the examples in the paper, the glasses vs. hat scenario, the listener in the speaker-listener pair is a “utility-maximizing agent,” assuming that the speaker is giving relevant and concise information. Because the speaker expresses information about the friend with glasses in a world comprised of people with glasses and people with glasses and a hat, the listener can assume that the speaker meant the friend with solely glasses because the other constituent possesses a supplemental detail that could have been stated in the description. This model attempts to target empirical assumptions, hyperbole, vague statements and more complex phenomena that the human brain is ambiguously trained to recognize despite lack of concrete evidence. The model takes a large leap towards future linguistic data and breakthrough; however, does not quite capture the holistic understanding of language. RSA may have the ability to be combined with other models in the future or improve enough to learn on pragmatic inferences in the future.