Prosody and performance in language production

Language production theories should explain how speakers generate an utterance's sound structure. One critical question is whether prosody and performance effects have different sources in the production system. It is argued that algorithms designed to predict phenomena such as pauses or intonational breaks are problematic because they tend to conflate prosody and planning. In addition, algorithms that have been proposed have not been evaluated systematically enough to allow their strengths and weaknesses to be assessed and compared, and to the extent that the algorithms have been evaluated, it is clear that they are only moderately successful at predicting the dependent measures of interest. Experimental work suggests that prosodic effects are based on prosodic constituency to the left of a potential boundary, and hesitations are due to planning of syntactic and semantic constituents to the right. Thus, any adequate algorithm must distinguish between prosody and performance, prosodic and syntactic-semantic constituency, and planning and execution effects. Notes 1Metrical trees (Liberman & Prince, Citation1977) serve a similar purpose. 2Specifically, the GG algorithm was designed to predict the duration of pauses 200 ms or longer at every sentential location. Nonetheless, in Ferreira (Citation1988) I argue that the algorithm can also be interpreted as predicting the probability of a pause, which in some circumstances makes the algorithm more plausible, since I found that people rarely paused more than once or twice in a single utterance, unless they were speaking at an unusually slow speech rate. 3The data were also analysed using the GG 200 ms the cutoff point, and results were similar. 4No intonational breaks were perceived for any token. 5I am grateful to an anonymous reviewer for making this point. 6The shortcomings of reading tasks that will be identified are true also of the version of the reading task created by Watson and Gibson (2004). 7Memorisation times were also recorded, but did not vary systematically with the independent variables (see Ferreira, 1991). 8The units could also be described as semantic, because the subject-predicate boundary is also a major semantic division. 9Ferreira (Citation1993) describes in detail in what way lengthening and pausing co-occur, and provides a mathematical model describing the trade-off between the two processes for implementing a metrical grid.