Detailed project information

Despite their native competence, speakers often make performance errors during language production. Do these errors influence communication, comprehension, language learning and language change?

Optimality Theory (OT) models competence, the static knowledge of language, while the Simulated Annealing for Optimality Theory Algorithm (SA-OT) imitates performance, the dynamic language production. SA-OT makes "errors": it returns ungrammatical forms violating the grammar (OT) but nevertheless uttered by speakers. The performance pattern also observable empirically emerges from the structure of SA-OT.

Speakers accept making errors in order to speak faster and more efficiently. The project asks how these errors affect communication. Computer simulations will introduce a virtual community of agents, with OT and SA-OT modelling their competence and performance respectively. Including a separate, full-fledged model of performance is a novelty in theoretical linguistics. Algorithms will be developed to realize comprehension and language acquisition.

By employing algorithms for comprehension, the project aims at understanding the conditions that make communication efficient: that permit a high production speed while the rate of misunderstanding remains low. Language evolution goes towards higher efficiency. Concurrently, the learning algorithms explain how the younger generations acquire their competence while having access exclusively to the teachers' (partially erroneous) performance. The learning algorithms should be good enough to enable efficient communication between subsequent generations. Yet, the not exactly perfect learning becomes a second driving force of language change.

The project will not only develop a theoretical model, but its predictions will be tested on concrete linguistic phenomena, vowel systems and double negation. Moreover, the results will be applied to improve speech technology. Speech synthesis turns more natural by introducing minor "errors", and speech recognition may benefit from understanding better the nature of errors in communication. Finally, the software tools developed during the project will be also made available to the community of linguists.