Speech Production

Notes on Language Systems and Language Production

C. Perfetti 2-8-00

The traditional view of language structure emphasizes levels of language description and assumes that these levels must contribute to the perception and production of language. The research question is how are they organized such that processing of language occurs the way it does.

Interpretation of syntactic structures
Concepts
Of course, connections (interfaces) between levels are critical. For example, Morpho-phonology, the interface between phonology and grammar: In English, ray+[pl]->rez; compare with race (s not z). Jumped->jumpt, mopt, (but moped) dragd (but ragged). And semantic relational concepts (actor, recipient) map onto syntactic structures (arguments and phrasal heads [subject, object])
In processing accounts, a main question has been constraints on interactions among the levels. Autonomies? Full interactions?
What does Jackendoff (1999) add to this account?

Basically, J affirms the foundations of language. The main foundation is Combinatoriality. This is what others have called productivity. The important implication is that a speaker's brain must have a lexicon and a set of combination principles (grammar).
J notes that Interfaces are required

Syntax->Phonology
Semantics (what J calls Conceptual Structure) ->Syntax
J's proposal is fully compatible with the traditional view. There is little new here, although J has made serious contributions to the semantics-syntax interface elsewhere. His rejection of a distinction between linguistic and conceptual semantics is accepted by many.
The most interesting aspect of J's treatment is the explicit expression of the Tripartite Parallel Architecture. This architecture seems to obvious to be radical. It could be if the ideas of "parallel" and "interactive" were specified.
In general the basic framework with an incorporated lexicon seems useful
Representational Modularity: Compare with Fodor. It has some differences that make it more cognitive and more formal. Languages of the Mind: A twist on propositional thought.
Implications for neuroscience: The Binding Problem

It is difficult, isn't it, to look at a speech production model such as Levelt's and wonder how in the world anyone manages to say anything.
Speech Production is hard, and one might make a case that it is the most challenging problem in psycholinguistics and probably in all the cognitive sciences. No wonder there are relatively few brave enough to work on it.
The phylogenetic and ontogenetic perspectives

The ontogenetic one is especially useful. If infants start with a nonphonologized set of syllables, they begin to phonologize under pressure from communication?
The Blueprint of the Speaker

Discourse/pragmatic considerations often ignored but L has a useful focus on them. It is less clear what happens to them once a speech plan is closer to being produced.
Grammatical Encoding

<select> + <us> ->selec tus
the importance of the phonological word form (experiments on homophone frequency); the access phase
the segments phase (speech errors constrain thinking here)
incremental generation of syllabic and metrical structure of the word
Utterance prosody

phonological phrases
intonational phrases
Phonetic encoding and articulation: The phonological score
Syllables can be stored gestures. Frequency statistics suggest 200 might be stored even in English

Traditionally support for these ideas-which are shared in general terms across several theories, especially M. Garrett, has come from speech errors. But there are some interesting experimental results also:
Lemmas Come first, before phonology

Picture interference studies. For example Schriefers, Meyer, & Levelt (1990, Journal of Memory and Language) presented subjects with line drawings to name and provided auditory interference (Other studies have used visual interference). For example, asked to name a picture of a dog, subjects might hear CAT, FOG, and various controls, including silence. Semantic interference occurred at -150ms ; phonological effects appeared at +150 ms and were facilitative. These effects are not obtained when Ss judge the pictures as old or new.

TOT study. Subjects were given a definition and asked to write down the word. Word were abstract nouns and objects. If they could not come up with it, they responded to several questions about phonological form and gender, confidence that they knew the word. They then were given the word.
Strong result: Subjects could often indicated gender of the word without any form information (e.g. number of syllables)

li libro, la penna
la prognosi, lo sperone: subset were like this, no cue to gender in the ending. Same result.

ERP study of subjects making decisions about pictures. Data suggest that grammatical information-Dutch gender-is available 40 ms before phonological information about word beginnings.