Syntactic Surprisal Affects Spoken Word Duration in Conversational Contexts

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 356–367, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics Syntactic surprisal affects spoken word duration in conversational contexts Vera Demberg, Asad B. Sayeed, Philip J. Gorinski, and Nikolaos Engonopoulos M2CI Cluster of Excellence and Department of Computational Linguistics and Phonetics Saarland University 66143 Saarbr¨ucken, Germany {vera,asayeed,philipg,nikolaos}@coli.uni-saarland.de Abstract We present results of a novel experiment to in- vestigate speech production in conversational data that links speech rate to information den- sity. We provide the first evidence for an asso- ciation between syntactic surprisal and word duration in recorded speech. Using the AMI corpus which contains transcriptions of focus group meetings with precise word durations, we show that word durations correlate with syntactic surprisal estimated from the incre- mental Roark parser over and above simpler measures, such as word duration estimated from a state-of-the-art text-to-speech system and word frequencies, and that the syntac- tic surprisal estimates are better predictors of word durations than a simpler version of sur- prisal based on trigram probabilities. This re- sult supports the uniform information density (UID) hypothesis and points a way to more re- alistic artificial speech generation. 1 Introduction The uniform information density (UID) hypothesis suggests that speakers try to distribute information uniformly across their utterances (Frank and Jaeger, 2008). Information density can be measured in terms of the surprisal incurred at each word, where surprisal is defined as the negative log-probability of an event. This paper sets out to test whether UID holds across different linguistic levels, i.e. whether speakers adapt word duration during production to syntactic surprisal, such that words with higher sur- prisal have longer durations than words with lower surprisal. We investigate this question in a corpus of transcribed speech from a mix of native and non- native English speakers, a population that is a non- trivial component of the user base for language tech- nologies developed for English. This data reflects a casual, uncontrolled conversational environment. Using linear mixed-effects modeling, we found that syntactic surprisal as calculated from a top- down incremental PCFG parser accounts for a sig- nificant amount of variation in spoken word dura- tion, using an HMM-trained text-to-speech system as a baseline. The findings of this paper provide ad- ditional support the uniform information density hy- pothesis and furthermore have implications for the design of text-to-speech systems, which currently do not take into account higher-level linguistic in- formation such as syntactic surprisal (or even word frequencies) for their word duration models. 1.1 Related work The use of word-level surprisal as a predictor of pro- cessing difficulty is based on the notion that pro- cessing difficulty results when a word is encountered that is unexpected given its preceding context. The amount of surprisal on a word wi can be formal- ized as the log of the inverse conditional probabil- ity of wi given the preceding words in the sentence w1 . . . wi−1, or − …

Content truncated to 500 words