Gregg C. Oden

Evaluation and integration of acoustic features in speech perception

Identification of synthetic stop consonants as either /bae/, /pae/, /dae/, or /tae/ was examined in two experiments in which the stimuli varied independently on voice onset time (VOT), the consonantal second and third formant (F2-F3) transitions and, in experiment 2, the intensity of the aspiration noise during the VOT period. In both experiments, the patterns of the resulting identification probabilities were complex, but systematic, functions of each of the independent variables. Of most interest was the fact that the likelihood of identifying a stimulus to be /bae/ or /pae/, rather than /dae/ or /tae/, was strongly influenced by the VOT as well as by the F2-F3 transitions. Analogously, the likelihood of identifying a stimulus to be /bae/ or /dae/, rather than /pae/ or /tae/, depended on the F2-F3 transitions as well as on VOT. Three explanations of these results were considered within a fuzzy logical model of speech perception: (1) that there is interaction in the evaluation of acoustic features, (2) that the listener requires more extreme values of acoustic features for some speech sounds than for that of other speech sounds, and (3) that the aspiration noise during the VOT period serves as an independent acoustic feature to distinguish /pae/ and /bae/ from /tae/ and /dae/.

Influence of context on the activation and selection of ambiguous word senses

The activation of ambiguous word senses was investigated by measuring the amount of interference in naming the ink color of a word that was either related or unrelated to one of the meanings of a preceding ambiguous word. In agreement with previous results obtained using this procedure (Conrad, 1974), evidence was obtained that both meanings of the ambiguous words are activated even in the presence of biasing context. However, contrary to previous findings, the degree of activation of each word sense depended on its degree of compatibility with the context. These results are consistent with a language processing system in which all interpretations of an ambiguity are accessed and then processed until an accurate determination has been made of which interpretation best satisfies the syntactic and semantic constraints that govern it.

Speech Perception: A Framework for Research and Theory

Publisher Summary This chapter discusses a psychological framework for speech perception. Scientists sometimes overcome complexity by analyzing an unmanageable problem into manageable subproblems. A naive speech perception model appears to give a reasonable description of the various processes and structures involved in a number of important phenomena. Faced with a speech stimulus, the listener must evaluate features of the sensory input and match these to pattern descriptions in memory. In accounting for certain phenomena, it is important to assume that the listener has information about the degree to which each feature is present in the speech sound. Similarly, the complexity of the sensory input argues for propositional descriptions of speech patterns and corresponding logical integration rules. This general framework allows for a natural description of the powerful contribution of both auditory and linguistic context. Whether or not this model proves to be ultimately correct, it should certainly serve as a valuable heuristic framework for research and theory in speech perception.

A fuzzy logical model of letter identification

Stimuli were generated by factorially varying two sets of features that distinguish between two letter patterns. Subjects rated the degree to which each stimulus was an instance of one letter rather than the alternative. The obtained ratings were relatively continuous and systematic functions of the feature manipulations. The results were well accounted for by a model in which (a) each feature has an associated fuzzy predicate that is used to independently evaluate the degree to which it is true that the feature is present in the stimulus; (b) the featural truth values are integrated according to fuzzy logical expressions that correspond directly to propositional descriptions of each letter pattern; and (c) the resulting goodness of match to the stimulus for each letter is compared to that of the alternatives to determine the final identification.

Fuzziness in semantic memory: Choosing exemplars of subjective categories.

Class membership is a fundamental relationship between concepts in semantic memory. Recent research indicates that, class membership may subjectively be a continuous type of relationship. The processing of information about the degree to which items belong to a particular class was investigated in an experiment in which subjects compared two statements describing class membership relationships. The results strongly supported a simple model which describes the judgment process as directly involving subjective degree-of-truthfulness values. The success of the model indicates that the subjects were able to process this kind of fuzzy information in a consistent and systematic manner. Some of the implications of the human competancy for processing fuzzy information are discussed.

Semantic constraints and judged preference for interpretations of ambiguous sentences

It is proposed that the degree of sensibleness of sentences is determined by semantic constraints which may be more or less satisfied. Such continuous semantic constraints were examined in two experiments in which subjects judged the likelihood of obtaining each of the interpretations of ambiguous sentences. The sentences were factorially generated by independently varying the degree to which semantic constraints for each interpretation were satisfied. In one experiment, the semantic constraints were manipulated by varying critical words within the ambiguous sentence; in the other experiment, a preceding context sentence was used. The results of both experiments supported the hypotheses that the judged likelihood was a direct function of the relative sensibleness of the interpretations, that semantic constraints determined the degree of sensibleness of each interpretation, and that these semantic constraints are continuous restrictions which are independent of each other and stable from sentence to sentence in which they occur.

Integration of fuzzy linguistic information in language comprehension

Abstract To successfully understand a natural language utterance, a person must simultaneously consider information about its perceptual, syntactic, semantic, and pragmatic structure. At each of these levels of analysis, the linguistic information is continuously but systematically variable. It is argued that in order to cope with and, in fact, exploit this continuous information requires fuzzy propositional representations of linguistic knowledge. Such a fuzzy propositional theoretical framework is proposed and models of several specific language processing components are developed and empirically evaluated within this framework. The model system incorporates many of the major principles of current cognitive psychological theory and extends them into the fuzzy domain.

INDEPENDENCE OF LEXICAL CONTEXT AND PHONOLOGICAL INFORMATION IN SPEECH PERCEPTION

M. A. Pitt (1995a) studied the joint influence of phonological information and lexical context in W. F. Ganongs (1980) task. Pitt improved on earlier studies by collecting enough observations to make possible the quantitative analyses of an individuals data. The present article shows that the results of such analyses demonstrate that the integration of phonological information and lexical context is very well accounted for by the fuzzy logical model of perception (FLMP). Although Pitt concluded that the results of his research argued against the FLMP in favor of an interactive feedback system, his conclusion was based on an analysis of transformed results. It is argued that this use of a response transformation led to incorrect conclusions and that ultimately, models must be tested directly against observed behavior.

The integration of contextual and featural information during word identification

Abstract The influence of featural and semantic information on word identification was examined in two experiments. Letter features were manipulated within words that occurred in sentence contexts supporting different interpretations to various degrees. In the first experiment, subjects were presented with sentences and asked to choose which interpretation of the word had been in the sentence. In the second experiment, the subject read each sentence aloud, allowing for the determination of how the critical word was identified. The combined results of the two experiments were taken to indicate that semantic and featural information jointly influence word identification, and that the obtained effect of context is unlikely to be due to factors that exert their influence after word identification has taken place. A fuzzy propositional model was used to provide an account of the results. In this model, the featural support for each interpretation of the stimulus is evaluated independently of contextual information, although contextual information influences the selection of the words identity. This illustrates that contextual information need not influence the sensory analysis of a stimulus in order to influence the identification of that stimulus.

Dependence, independence, and emergence of word features.

The degree to which a letter is perceived to be an e rather than a c is a function of the central, horizontal bar, which can be made to be more or less present in any given stimulus. By this means, word-form stimuli were produced through the independent, continuous manipulation of e versus c and of r versus h in two word frames (wat-- and --ase) yielding stimuli that ranged between water and watch in the former instance and between erase and chase in the latter. The identification probabilities for these stimuli were well accounted for in each case by a version of a fuzzy propositional model that is based on the assumption of independent features, indicating that the manipulated features within each word frame were not perceptually interdependent. However, the probabilities differed systematically between the two stimulus matrices; this means that there must be some other kind of higher order effect involved in identification. The hypothesis that the interaction was due to an emergent word envelope feature was rejected on the basis of the model analysis.