Models Of Spoken Word Recognition Research Articles

Orthographic Influences When Processing Spoken Pseudowords: Theoretical Implications

When we hear an utterance, is the orthographic representation of that utterance activated when it is being processed? Orthographic influences have been previously examined in relation to spoken pseudoword processing in three different paradigms. Unlike real word processing, no orthographic effects with pseudowords have been observed in a phoneme goodness ratings task, and there is a mixed outcome in studies looking for spelling–sound consistency effects. In contrast, the orthography of spoken pseudohomographs has been shown to be activated, given that they prime their homographic base word. Explanations are sought for the findings in these three paradigms, leading to an exploration of theoretical models of spoken word recognition.

Experimental evidence for phoneme‐level inhibition in spoken word recognition.

The TRACE model of spoken word recognition [McClelland and Elman (1986)] contains phoneme‐level inhibition whereas the MERGE model [Norris et al. (2000)] does not. Little has been reported in the experimental literature to bear on the question of phoneme‐level inhibition, so the current study aims to fill this gap. An auditory lexical decision task was conducted, where participants were tested on English words preceded by isolated English sounds as primes. The relatedness of the prime sound to the final sound of the target word was manipulated. Phoneme‐level inhibition is demonstrated when participants respond more slowly in the Similar condition (e.g., /sh/ before “bus”) than in the Identical (e.g., /s/ before “bus”) and Unrelated (e.g., /m/ before “bus”) conditions. This task revealed that phoneme‐level inhibition was active at ISIs of 250 and 500 ms, but not 50 ms. A second lexical decision task indicated that inhibition begins after 125 ms and ends by 1000 ms. These results support models of word recognition that contain phoneme‐level inhibition. Finally, a single‐word shadowing task, in which participants repeat words primed by isolated English sounds as quickly as possible, is expected to replicate these findings and to extend them by adding speech error data.

Clustering coefficients of lexical neighborhoods: Does neighborhood structure matter in spoken word recognition?

High neighborhood density reduces the speed and accuracy of spoken word recognition. The two studies reported here investigated whether Clustering Coefficient (CC) - a graph theoretic variable measuring the degree to which a word's neighbors are neighbors of one another, has similar effects on spoken word recognition. In Experiment 1, we found that high CC words were identified less accurately when spectrally degraded than low CC words. In Experiment 2, using a word repetition procedure, we observed longer response latencies for high CC words compared to low CC words. Taken together, the results of both studies indicate that higher CC leads to slower and less accurate spoken word recognition. The results are discussed in terms of activation-plus-competition models of spoken word recognition.

The roles of tonal and segmental information in Mandarin spoken word recognition: An eyetracking study

We used eyetracking to examine how tonal versus segmental information influence spoken word recognition in Mandarin Chinese. Participants heard an auditory word and were required to identify its corresponding picture from an array that included the target item ( chuang2 ‘bed’), a phonological competitor (segmental: chuang1 ‘window’; cohort: chuan2 ‘ship’; rhyme: huang2 ‘yellow’; tonal: niu2 ‘cow’), and two phonologically unrelated distractors. Growth curve analysis was used to characterize the trajectory of looks to target and competitor items during word processing. We found similar model fits for the segmental and cohort conditions characterized by slower eye movements to correct targets compared to baseline, suggesting that tonal and segmental information are accessed concurrently and play comparable roles in constraining activation. These findings are discussed with respect to current models of spoken word recognition that have not previously accounted for the role of tone.

Looking for phoneme-level inhibition in spoken word recognition using auditory lexical decision.

The TRACE model of spoken word recognition [McClelland & Elman (1986)] contains phoneme-level inhibition while the MERGE model [Norris et al. (2000)] does not include active competition between phonemes. Previous work found evidence for facilitation at the phoneme level and competition at the lexical level, but little research has been conducted looking for the existence of inhibition between phonemes. Using an auditory lexical decision task, the current study looks for phoneme-level inhibition by giving participants English target words preceded by isolated English sounds as primes. There are three conditions depending on how related the prime sound is to the final phoneme of the target word: identical (e.g., [s] before “gas”), similar (e.g., [s] before “cash”), and unrelated (e.g., [m] before gas). Facilitation, or faster average reaction times (RTs), is expected in the identical condition relative to the baseline unrelated condition. If facilitation is also found in the similar condition, the results will provide evidence against phoneme-level inhibition, supporting an approach with gradient activation of phonemes. However, if the average RT for the similar condition is greater than or equal to that of the unrelated condition, the results will support models of spoken word recognition containing phoneme-level inhibition.

The influence of the phonological neighborhood clustering coefficient on spoken word recognition.

Clustering coefficient-a measure derived from the new science of networks-refers to the proportion of phonological neighbors of a target word that are also neighbors of each other. Consider the words bat, hat, and can, all of which are neighbors of the word cat; the words bat and hat are also neighbors of each other. In a perceptual identification task, words with a low clustering coefficient (i.e., few neighbors are neighbors of each other) were more accurately identified than words with a high clustering coefficient (i.e., many neighbors are neighbors of each other). In a lexical decision task, words with a low clustering coefficient were responded to more quickly than words with a high clustering coefficient. These findings suggest that the structure of the lexicon (i.e., the similarity relationships among neighbors of the target word measured by clustering coefficient) influences lexical access in spoken word recognition. Simulations of the TRACE and Shortlist models of spoken word recognition failed to account for the present findings. A framework for a new model of spoken word recognition is proposed.

Pre-lexical abstraction of speech in the auditory cortex

Speech perception requires the decoding of complex acoustic patterns. According to most cognitive models of spoken word recognition, this complexity is dealt with before lexical access via a process of abstraction from the acoustic signal to pre-lexical categories. It is currently unclear how these categories are implemented in the auditory cortex. Recent advances in animal neurophysiology and human functional imaging have made it possible to investigate the processing of speech in terms of probabilistic cortical maps rather than simple cognitive subtraction, which will enable us to relate neurometric data more directly to behavioural studies. We suggest that integration of insights from cognitive science, neurophysiology and functional imaging is necessary for furthering our understanding of pre-lexical abstraction in the cortex.

Word Length and Lexical Competition: Longer is the Same as Shorter

Neighborhood density refers to the number of words that sound similar to a given word. Previous studies have found that neighborhood density influences the recognition of spoken words (Luce & Pisoni, 1998); however, this work has focused almost exclusively on monosyllabic words in English. To investigate the effects of neighborhood density on longer words, bisyllabic words varying in neighborhood density were presented auditorily to participants in a perceptual identification task and a lexical decision task. In the perceptual identification task, words with sparse neighborhoods were more accurately identified than words with dense neighborhoods. In the lexical decision task, words with sparse neighborhoods were responded to more quickly and more accurately than words with dense neighborhoods. These results are similar to those found in studies examining the influence of neighborhood density on the recognition of monosyllabic words in English. In order to better understand lexical processing, models of spoken word recognition must account for the processing of words of all types.

Within-category VOT affects recovery from “lexical” garden-paths: Evidence against phoneme-level inhibition

Spoken word recognition shows gradient sensitivity to within-category voice onset time (VOT), as predicted by several current models of spoken word recognition, including TRACE (McClelland, J., & Elman, J. (1986). The TRACE model of speech perception. Cognitive Psychology, 18, 1–86). It remains unclear, however, whether this sensitivity is short-lived or whether it persists over multiple syllables. VOT continua were synthesized for pairs of words like barricade and parakeet, which differ in the voicing of their initial phoneme, but otherwise overlap for at least four phonemes, creating an opportunity for “lexical garden-paths” when listeners encounter the phonemic information consistent with only one member of the pair. Simulations established that phoneme-level inhibition in TRACE eliminates sensitivity to VOT too rapidly to influence recovery. However, in two Visual World experiments, look-contingent and response-contingent analyses demonstrated effects of word initial VOT on lexical garden-path recovery. These results are inconsistent with inhibition at the phoneme level and support models of spoken word recognition in which sub-phonetic detail is preserved throughout the processing system.

Feedback consistency effects in visual and auditory word recognition: Where do we stand after more than a decade?

The role of phonology-to-spelling consistency (i.e., feedback consistency) was investigated in 3 lexical decision experiments in both the visual and auditory modalities in French and English. No evidence for a feedback consistency effect was found in the visual modality, either in English or in French, despite the fact that consistency was manipulated for different kinds of units (onsets and rimes). In contrast, robust feedback consistency effects were obtained in the auditory lexical decision task in both English and French when exactly the same items that produced a null effect in the visual modality were used. Neural network simulations are presented to show that previous demonstrations of feedback consistency effects in the visual modality can be simulated with a model that is not sensitive to feedback consistency, suggesting that these effects might have come from various confounds. These simulations, together with the authors' results, suggest that there are no feedback consistency effects in the visual modality. In contrast, such effects are clearly present in the auditory modality. Given that orthographic information is absent from current models of spoken word recognition, the present findings present a major challenge to these models.

JTRACE: A reimplementation and extension of the TRACE model of speech perception and spoken word recognition

This article describes jTRACE, a freely available, cross-platform Java reimplementation of the TRACE model of spoken word recognition. The goal of the reimplementation is to facilitate the use of simulations by researchers who may not have the skills or time necessary to use or extend the original C implementation. In this article, we report a large-scale validation project, in which we have replicated a number of important previous simulations, and then we describe several new features in jTRACE designed to help researchers conduct original TRACE research, as well as to replicate earlier findings. These features include visualization tools, powerful scripting, built-in data graphing, adjustable levels of external and internal noise, and adjustable lexical characteristics, such as frequency of occurrence. Functions for saving and reloading entire simulations facilitate archiving, sharing, and replication and also make jTRACE ideal for educational use, since it comes bundled with several important simulations. jTRACE can be downloaded from magnuson.psy.uconn.edu/jtrace.

The spread of the phonological neighborhood influences spoken word recognition

In three experiments, the processing of words that had the same overall number of neighbors but varied in the spread of the neighborhood (i.e., the number of individual phonemes that could be changed to form real words) was examined. In an auditory lexical decision task, a naming task, and a same-different task, words in which changes at only two phoneme positions formed neighbors were responded to more quickly than words in which changes at all three phoneme positions formed neighbors. Additional analyses ruled out an account based on the computationally derived uniqueness points of the words. Although previous studies (e.g., Luce & Pisoni, 1998) have shown that the number of phonological neighbors influences spoken word recognition, the present results show that the nature of the relationship of the neighbors to the target word--as measured by the spread of the neighborhood--also influences spoken word recognition. The implications of this result for models of spoken word recognition are discussed.

JT R A C E: A user-friendly reimplementation and extension of the T R A C E model of speech perception and spoken word recognition

jTRACE is a cross-platform Java reimplementation and extension of the TRACE model of spoken word recognition. TRACE (McClelland and Elman, 1986) is a three-layer interactive-activation model of speech perception that takes a pseudoacoustic input representation and accurately simulates many phenomena of human speech processing. The goals of the reimplementation are to provide a user-friendly version of the model that can run on any current computer platform, extend the model with graphing and scripting tools, and thus provide a tool that is powerful enough for intensive research while easy enough to use in a classroom setting. New features in jTRACE include simulation visualization, results graphing, powerful scripting, internal and external noise parameters, adjustable lexical and phonological characteristics, word recognition metrics, data exporting, and saving/archiving tools. A set of historically relevant simulations is bundled with the program. This feature is ideal for educational applications. Any simulation can be saved in a similar format, promoting sharing and replication. jTRACE was validated quantitatively against the original implementation by replicating key simulations reported in the literature. jTRACE can be downloaded freely from <http://maglab.psy.uconn.edu/jtrace.html> [Work supported by National Institute on Deafness and Other Communication Disorders.]

Word length and lexical activation: Longer is better.

Many models of spoken word recognition posit the existence of lexical and sublexical representations, with excitatory and inhibitory mechanisms used to affect the activation levels of such representations. Bottom-up evidence provides excitatory input, and inhibition from phonetically similar representations leads to lexical competition. In such a system, long words should produce stronger lexical activation than short words, for 2 reasons: Long words provide more bottom-up evidence than short words, and short words are subject to greater inhibition due to the existence of more similar words. Four experiments provide evidence for this view. In addition, reaction-time-based partitioning of the data shows that long words generate greater activation that is available both earlier and for a longer time than is the case for short words. As a result, lexical influences on phoneme identification are extremely robust for long words but are quite fragile and condition-dependent for short words. Models of word recognition must consider words of all lengths to capture the true dynamics of lexical activation.

Lexical boosting of noise-band speech in open- and closed-set formats

Recognition of spoken words in noise and in quiet is more accurate for Lexically Easy words (high frequency words with few similar-sounding neighbors) than for Lexically Hard words (low frequency words with many similar-sounding neighbors). Using monosyllables, the present set of two experiments extends this finding to a perceptually interesting class of stimuli and test formats. In both open- and closed-sets, listeners attempted to identify amplitude-modulated and bandpass-filtered words [Shannon, R., Zeng, F., Kamath, V., Wgonski, J., Ekelid, M., 1995. Speech recognition with primarily temporal cues. Science 270, 303–304]––noise-band speech––shown to simulate the performance of cochlear implant (CI) patients using the same number of frequency channels. The words were synthesized from a database that controls for Lexical Difficulty, Talker Identity and Talker Gender. Word recognition was significantly more accurate for Easy words in both the open- and the closed-set experiments. These results indicate that, even when spoken word recognition is challenged by noise-band speech, the Easy–Hard effect survives the perceptually uncertain conditions of word variability. Consequences for models of spoken word recognition are explored.

Neighborhood density effects in spoken word recognition in Spanish

The present work examined the relationships among familiarity ratings, frequency of occurrence, neighborhood density, and word length in a corpus of Spanish words. The observed relationships were similar to the relationships found among the same variables in English. An auditory lexical decision task was then performed to examine the influence of word frequency, neighborhood density, and neighborhood frequency on spoken word recognition in Spanish. In contrast to the competitive effect of phonological neighborhoods typically observed in English, a facilitative effect of neighborhood density and neighborhood frequency was found in Spanish. Implications for models of spoken word recognition and language disorders are discussed.

Rapid adaptation to foreign-accented English

This study explored the perceptual benefits of brief exposure to non-native speech. Native English listeners were exposed to English sentences produced by non-native speakers. Perceptual processing speed was tracked by measuring reaction times to visual probe words following each sentence. Three experiments using Spanish- and Chinese-accented speech indicate that processing speed is initially slower for accented speech than for native speech but that this deficit diminishes within one minute of exposure. Control conditions rule out explanations for the adaptation effect based on practice with the task and general strategies for dealing with difficult speech. Further results suggest that adaptation can occur within as few as two to four sentence-length utterances. The findings emphasize the flexibility of human speech processing and require models of spoken word recognition that can rapidly accommodate significant acoustic-phonetic deviations from native language speech patterns.

Frequency effects in phoneme processing

This study tested the hypothesis that phonological segments are activated during word recognition in proportion to their frequency of use, analogous to frequency effects in whole word recognition. Preliminary evidence for the hypothesis was given by Moates, Bond, and Stockmal [LabPhon 7 (2002)] using a word reconstruction task. The present study used a gating task in which progressively longer fragments of a word are presented to listeners who must identify the word after as few gates as possible. High- and low-frequency segments were contrasted by presenting them in word pairs that differed in two segments, e.g., collision–collusion, where /I/ is more frequent than /u/. We constructed 15 words pairs contrasting vowels and 16 pairs contrasting consonants (e.g., relief–release, where /s/ is more frequent than /f/). Identification judgments were gathered from 125 participants. An ANOVA showed high-frequency consonants to be identified at significantly earlier gates than their matched low-frequency consonants with both subjects and items as random factors. No such effect appeared for vowels. Also, whole words containing the high-frequency segments were not identified significantly earlier than those containing low-frequency segments. If the phoneme frequency effect is reliable, then spoken word recognition models should address it.

Fundamentals of spoken word recognition

Researchers have made significant progress in identifying the basic principles responsible for the normal-listener’s rapid and accurate identification of spoken words. In particular, there is now almost uniform consensus that spoken word recognition involves two fundamental processes: activation and competition. Most current models of recognition propose that stimulus input (i.e., a spoken word) activates a set of representations of similar sounding words in memory that subsequently vie for recognition. Despite the fact that similarity is afforded a crucial role in activating and discriminating among lexical competitors, we currently have little precise information regarding perceived similarity relations among spoken words. We will discuss the fundamental role of similarity in current models of spoken word recognition, past attempts to capture perceived similarity among spoken words, and ongoing efforts in our laboratory to understand more deeply the precise role of similarity in the perception of spoken stimuli. [Work supported by NIH.]

Flow of information in the spoken word recognition system

Spoken word recognition consists of two major component processes. First, at the prelexical stage, an abstract description of the utterance is generated from the information in the speech signal. Second, at the lexical stage, this description is used to activate all the words stored in the mental lexicon which match the input. These multiple candidate words then compete with each other. We review evidence which suggests that positive (match) and negative (mismatch) information of both a segmental and a suprasegmental nature is used to constrain this activation and competition process. We then ask whether, in addition to the necessary influence of the prelexical stage on the lexical stage, there is also feedback from the lexicon to the prelexical level. In two phonetic categorization experiments, Dutch listeners were asked to label both syllable-initial and syllable-final ambiguous fricatives (e.g., sounds ranging from [f] to [s]) in the word–nonword series maf–mas, and the nonword–word series jaf–jas. They tended to label the sounds in a lexically consistent manner (i.e., consistent with the word endpoints of the series). These lexical effects became smaller in listeners’ slower responses, even when the listeners were put under pressure to respond as fast as possible. Our results challenge models of spoken word recognition in which feedback modulates the prelexical analysis of the component sounds of a word whenever that word is heard.