Models Of Spoken Word Recognition Research Articles

Phonological neighbors cooperate during spoken-sentence processing: Evidence from a nonword detection task.

We used a novel nonword detection task to examine the lexical competition principle postulated in most models of spoken word recognition. To do so, in Experiment 1 we presented sequences of spoken words with half of the sequences containing a nonword, and the target nonword (i.e., press a response key whenever you detect a nonword in the sequence) could either be phonologically related (a phonological neighbor) or unrelated to the immediately preceding word. We reasoned that the reactivation of a phonological neighbor during target nonword processing should delay the moment at which a nonword decision can be made. Contrary to our hypothesis, participants were faster at detecting nonwords when they were preceded by a phonological neighbor compared with an unrelated word. In Experiment 2, an inhibitory effect of phonological relatedness on nonword decisions was observed in a classic priming situation using the same set of related and unrelated word-nonword pairs. We discuss the implications of these findings in regard to the main models of spoken word recognition, and conclude that our specific experimental set-up with phonological neighbors embedded in spoken sentences is more sensitive to cooperative interactions between co-activated sublexical representations than lexical competition between co-activated lexical representations, with the latter being modulated by whether or not the words compete for the same slot in time.

Lexical Feedback in the Time-Invariant String Kernel (TISK) Model of Spoken Word Recognition.

The Time-Invariant String Kernel (TISK) model of spoken word recognition (Hannagan, Magnuson & Grainger, 2013; You & Magnuson, 2018) is an interactive activation model with many similarities to TRACE (McClelland & Elman, 1986). However, by replacing most time-specific nodes in TRACE with time-invariant open-diphone nodes, TISK uses orders of magnitude fewer nodes and connections than TRACE. Although TISK performed remarkably similarly to TRACE in simulations reported by Hannagan et al., the original TISK implementation did not include lexical feedback, precluding simulation of top-down effects, and leaving open the possibility that adding feedback to TISK might fundamentally alter its performance. Here, we demonstrate that when lexical feedback is added to TISK, it gains the ability to simulate top-down effects without losing the ability to simulate the fundamental phenomena tested by Hannagan et al. Furthermore, with feedback, TISK demonstrates graceful degradation when noise is added to input, although parameters can be found that also promote (less) graceful degradation without feedback. We review arguments for and against feedback in cognitive architectures, and conclude that feedback provides a computationally efficient basis for robust constraint-based processing.

Semantic richness effects in isolated spoken word recognition: Evidence from massive auditory lexical decision.

While known to influence visual lexical processing, the semantic information we associate with words has recently been found to influence auditory lexical processing as well. The present work explored the influence of semantic richness in auditory lexical decision. Study 1 recreated an experiment investigating semantic richness effects in concrete nouns (Goh et al., 2016). In Study 2, we expanded the stimulus set from 442 to 8,626 items, exploring the robustness of effects observed in Study 1 against a larger data set with increased diversity in both word class and other characteristics of interest. We also utilized generalized additive mixed models to investigate potential nonlinear effects. Results indicate that semantic richness effects become more nuanced and detectable when a wider set of items belonging to different parts of speech is examined. Findings are discussed in the context of models of spoken word recognition. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Closure duration versus f0 perturbation as a cue to underlying stops for the American English intervocalic flap

Researchers found that there are five main cues for distinguishing the voicing of these intervocalic plosives, preceding vowel duration, following vowel duration, closure duration, semantic context, and f0 perturbation, also known as consonant intrinsic f0. In American English, when /t/ and /d/ occur intervocalically, they are realized as a voiced alveolar flap [ɾ] (e.g., [ɹa͡ɪɾɪŋ] writing/riding, [liɾɚ] liter/leader). The presence of these cues indicates incomplete neutralization of these forms. To address a current gap in the literature, the present study focused on American English native speakers’ perception and use of CF0 and closure duration as independent and combined informative cues for deciding voicing contrast of /t/ and /d/ when neutralized by the flap. Participants engaged in a self-administered online forced-judgement task of a pseudoword in 30 combinations of CF0 and flap closure duration to indicate the underlying representation (/hɑtɑ/ or /hɑdɑ/) of the surface production [hɑɾɑ] they perceived. This study observed that participants are more likely to select /hɑtɑ/ when the CF0 is higher and / hɑdɑ/ elsewhere. These modest findings suggest that CF0 is a useful cue for voicing distinction. The results suggest that models of spoken word recognition and speech perception ought to include CF0 as a cue.

Swift Prosodic Modulation of Lexical Access: Brain Potentials From Three North Germanic Language Varieties.

According to most models of spoken word recognition, listeners probabilistically activate a set of lexical candidates, which is incrementally updated as the speech signal unfolds. Speech carries segmental (speech sound) as well as suprasegmental (prosodic) information. The role of the latter in spoken word recognition is less clear. We investigated how suprasegments (tone and voice quality) in three North Germanic language varieties affected lexical access by scrutinizing temporally fine-grained neurophysiological effects of lexical uncertainty and information gain. Three event-related potential (ERP) studies were reanalyzed. In all varieties investigated, suprasegments are associated with specific word endings. Swedish has two lexical "word accents" realized as pitch falls with different timings across dialects. In Danish, the distinction is in voice quality. We combined pronunciation lexica and frequency lists to calculate estimates of lexical uncertainty about an unfolding word and information gain upon hearing a suprasegmental cue and the segment upon which it manifests. We used single-trial mixed-effects regression models run every 4 ms. Only lexical uncertainty showed solid results: a frontal effect at 150-400 ms after suprasegmental cue onset and a later posterior effect after 200 ms. While a model including only segmental information mostly performed better, it was outperformed by the suprasegmental model at 200-330 ms at frontal sites. The study points to suprasegmental cues contributing to lexical access over and beyond segments after around 200 ms in the North Germanic varieties investigated. Furthermore, the findings indicate that a previously reported "pre-activation negativity" predominantly reflects forward-looking processing. https://doi.org/10.23641/asha.25016486.

Contra assertions, feedback improves word recognition: How feedback and lateral inhibition sharpen signals over noise

Whether top-down feedback modulates perception has deep implications for cognitive theories. Debate has been vigorous in the domain of spoken word recognition, where competing computational models and agreement on at least one diagnostic experimental paradigm suggest that the debate may eventually be resolvable. Norris and Cutler (2021) revisit arguments against lexical feedback in spoken word recognition models. They also incorrectly claim that recent computational demonstrations that feedback promotes accuracy and speed under noise (Magnuson et al., 2018) were due to the use of the Luce choice rule rather than adding noise to inputs (noise was in fact added directly to inputs). They also claim that feedback cannot improve word recognition because feedback cannot distinguish signal from noise. We have two goals in this paper. First, we correct the record about the simulations of Magnuson et al. (2018). Second, we explain how interactive activation models selectively sharpen signals via joint effects of feedback and lateral inhibition that boost lexically-coherent sublexical patterns over noise. We also review a growing body of behavioral and neural results consistent with feedback and inconsistent with autonomous (non-feedback) architectures, and conclude that parsimony supports feedback. We close by discussing the potential for synergy between autonomous and interactive approaches.

When phonological neighbours cooperate during spoken sentence processing.

This study examined for the first time the impact of the presence of a phonological neighbour on word recognition when the target word and its neighbour co-occur in a spoken sentence. To do so, we developed a new task, the verb detection task, in which participants were instructed to respond as soon as they detected a verb in a sequence of words, thus allowing us to probe spoken word recognition processes in real time. We found that participants were faster at detecting a verb when it was phonologically related to the preceding noun than when it was phonologically unrelated. This effect was found with both correct sentences (Experiment 1) and with ungrammatical sequences of words (Experiment 2). The effect was also found in Experiment 3 where adjacent phonologically related words were included in the non-verb condition (i.e., word sequences not containing a verb), thus ruling out any strategic influences. These results suggest that activation persists across different words during spoken sentence processing such that processing of a word at position n + 1 benefits from the sublexical phonology activated during processing of the word at position n. We discuss how different models of spoken word recognition might be able (or not) to account for these findings.

The activation of embedded (pseudo-)stems in auditory lexical processing: implications for models of spoken word recognition

ABSTRACT A large literature on visual word recognition has examined the role of (apparent) morphological structure by comparing suffixed (such as treatment), pseudo-suffixed (pigment), and non-suffixed (dogma) words with respect to their embeddings (treat, pig, dog). We examined the processing of these word types, as well as semantic controls, in an auditory primed lexical decision paradigm. The results show significant priming in all conditions relative to an unrelated baseline, with larger priming effects for truly suffixed words than for pseudo-suffixed and non-suffixed words. The results suggest that initial embeddings are activated in spoken word processing, and remain active in ways that do not depend on (apparent) morphological structure. We discuss the implications of these findings for models of lexical access that predict inhibition of disfavoured competitors and models that hold that attempted decomposition is driven by meaning relatedness between the carrier word and its possible embedded stem(s).

Joint, distributed and hierarchically organized encoding of linguistic features in the human auditory cortex.

The precise role of the human auditory cortex in representing speech sounds and transforming them to meaning is not yet fully understood. Here we used intracranial recordings from the auditory cortex of neurosurgical patients as they listened to natural speech. We found an explicit, temporally ordered and anatomically distributed neural encoding of multiple linguistic features, including phonetic, prelexical phonotactics, word frequency, and lexical-phonological and lexical-semantic information. Grouping neural sites on the basis of their encoded linguistic features revealed a hierarchical pattern, with distinct representations of prelexical and postlexical features distributed across various auditory areas. While sites with longer response latencies and greater distance from the primary auditory cortex encoded higher-level linguistic features, the encoding of lower-level features was preserved and not discarded. Our study reveals a cumulative mapping of sound to meaning and provides empirical evidence for validating neurolinguistic and psycholinguistic models of spoken word recognition that preserve the acoustic variations in speech.

Multilingual spoken word recognition: A megastudy approach

Bilingualism research has primarily focused on the perception and processing of individual sounds or word learning. Many studies have investigated how bilingual listeners perceive sound contrasts that don't create lexical distinctions in their native language. There is substantially less research that has investigated how word-level properties impact L2 auditory processing. The present study examines how auditory lexical processing differs between monolingual and bilingual listeners with different language backgrounds. We collected lexical decision accuracies and latencies for 26,793 words and 9600 pseudowords from the Massive Auditory Lexical Decision database from native and non-native listener responses. We compare the language backgrounds of our 1028 listeners and group them into four groups: native speakers, early, early-late, and late bilinguals. We report the findings of an analysis investigating how language background and proficiency modulate lexical effects to understand how language background influences spoken word recognition. We find differences in the effects of lexical frequency, phonological neighborhood density, and phonological uniqueness point between the different listener groups. We discuss the impact of bilingualism in a word recognition task and how these results inform models of spoken word recognition and second language acquisition.

Decoding syntactic class from EEG during spoken word recognition

A fundamental issue in spoken language comprehension involves understanding the interaction of linguistic representations across different levels of organization (e.g., phonological, lexical, syntactic, and semantic). In particular, there is debate about when different levels are accessed during spoken word recognition. Under serial processing models, comprehension is sequential. In contrast, under parallel processing models, simultaneous activation of representations at multiple levels can occur. The current study investigates this issue by isolating neural responses to syntactic class distinctions from acoustic and phonological responses. EEG data were collected in an event-related potential (ERP) experiment in which participants (N = 26) listened to words varying in syntactic class (nouns versus adjectives) that were controlled for low-level acoustic differences via cross-splicing. Machine learning techniques were used to decode syntactic class from ERP responses over time. Results showed that syntactic class is decodable approximately 160–190 ms after the average syntactic point of disambiguation in the words, during which listeners are still processing acoustic information. This supports the prediction that different levels of representation have overlapping timecourses. Overall, these results are consistent with a parallel, interactive processing model of spoken word recognition, in which higher-level information—such as syntactic class—is accessed while acoustic analysis is still occurring.

Computational Modeling of an Auditory Lexical Decision Experiment Using DIANA.

We present an implementation of DIANA, a computational model of spoken word recognition, to model responses collected in the Massive Auditory Lexical Decision (MALD) project. DIANA is an end-to-end model, including an activation and decision component that takes the acoustic signal as input, activates internal word representations, and outputs lexicality judgments and estimated response latencies. Simulation 1 presents the process of creating acoustic models required by DIANA to analyze novel speech input. Simulation 2 investigates DIANA's performance in determining whether the input signal is a word present in the lexicon or a pseudoword. In Simulation 3, we generate estimates of response latency and correlate them with general tendencies in participant responses in MALD data. We find that DIANA performs fairly well in free word recognition and lexical decision. However, the current approach for estimating response latency provides estimates opposite to those found in behavioral data. We discuss these findings and offer suggestions as to what a contemporary model of spoken word recognition should be able to do.

When facilitation becomes inhibition: effects of modality and lexicality on transposed-phoneme priming

ABSTRACT We examined the contributions of phoneme-to-word facilitation and word-to-word inhibition to transposed-phoneme priming effects under unimodal and cross-modal presentations. Experiments 1A and 1B showed that the presentation of an auditory prime formed by transposing two phonemes in a given target word facilitated lexical decisions to auditory targets. This facilitation was independent of the lexicality of the primes. In Experiment 2 the targets were presented visually rather than auditorily. We found an inhibitory priming effect, which, in contrast to Experiment 1, was influenced by the lexicality of the primes, with an effect emerging only with word primes. These findings point to a greater impact of phoneme-to-word facilitation under unimodal presentation and a greater role for word-to-word inhibition under cross-modal presentation. Hence, by simply manipulating the modality of target presentation, it is possible to separately probe two central mechanisms postulated in models of spoken word recognition, namely phoneme-to-word activation and lexical competition.

Perception and timing of acoustic distance

The notion of acoustic distance figures into many aspects of phonetics, including phonological neighborhoods. A measurement of word-level acoustic distance useful for cognitive modeling must account for two listener characteristics: sensitivity to acoustic differences and sensitivity to duration discrepancies between words. The present work used dynamic time warping to measure how acoustic distance accumulates between words over time. The results of a distance rating task with synthesized vowels were used as a basis for selecting a mathematical function that best matched listener sensitivities. Additionally, the results of a reminder task with synthesized vowels were used to determine a just noticeable difference threshold for vowel duration. The results suggested that a distance function based on the 4.5-norm using a 30 ms radius for dynamic time warping best matched human behavior. A third analysis used these dynamic time warping configurations to model reaction times in an auditory lexical decision task and found that Euclidean distance and no temporal constraints on dynamic time warping best matched human behavior. These results are discussed in relation to spoken word recognition models, including how to assess the acoustic match between the speech signal and a word in the lexicon.

The recognition of spoken pseudowords

ABSTRACT Pseudowords are used as stimuli in many psycholinguistic experiments, yet they remain largely under-researched. To better understand the cognitive processing of pseudowords, we analysed the pseudoword responses in the Massive Auditory Lexical Decision megastudy data set. Linguistic characteristics that influence the processing of real English words – namely, phonotactic probability, phonological neighbourhood density, uniqueness point, and morphological complexity – were also found to influence the processing time of spoken pseudowords. Subsequently, we analysed how the linguistic characteristics of non-unique portions of pseudowords influenced processing time. We again found that the named linguistic characteristics affected processing time, highlighting the dynamicity of activation and competition. We argue these findings also speak to learning new words and spoken word recognition generally. We then discuss what aspects of pseudoword recognition a full model of spoken word recognition must account for. We finish with a re-description of the auditory lexical decision task in light of our results.

When you hear /baksɛt/ do you think /baskɛt/? Evidence for transposed-phoneme effect with multisyllabic words.

In this study we asked whether nonwords created by transposing two phonemes (/biksɔt/) are perceived as being more similar to their base words (/biskɔt/) than nonwords created by substituting two phonemes (/bipfɔt/). Using the short-term phonological priming and a lexical-decision task, Experiment 1 showed that transposed-phoneme nonword primes lead to shorter RTs on the target base words than substituted-phoneme nonword primes. Using a single-presentation lexical-decision task, Experiment 2 showed that transposed-phoneme nonwords lead to longer "no" decision responses than substituted-phoneme nonwords. In both Experiments 1 and 2, the transposed-phoneme effect was observed when the transposed phonemes were adjacent (/biksɔt/-/biskɔt/) but not when they were distant (/ʃoloka/-/ʃokola/). Our findings suggest that nonwords created by transposing adjacent phonemes in real words generate more activation of the lexical representations associated with the base words than do matched control nonwords. More generally, our findings present a challenge for models of spoken word recognition that code for the precise order of speech segments. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

What Do Cognitive Networks Do? Simulations of Spoken Word Recognition Using the Cognitive Network Science Approach.

Cognitive network science is an emerging approach that uses the mathematical tools of network science to map the relationships among representations stored in memory to examine how that structure might influence processing. In the present study, we used computer simulations to compare the ability of a well-known model of spoken word recognition, TRACE, to the ability of a cognitive network model with a spreading activation-like process to account for the findings from several previously published behavioral studies of language processing. In all four simulations, the TRACE model failed to retrieve a sufficient number of words to assess if it could replicate the behavioral findings. The cognitive network model successfully replicated the behavioral findings in Simulations 1 and 2. However, in Simulation 3a, the cognitive network did not replicate the behavioral findings, perhaps because an additional mechanism was not implemented in the model. However, in Simulation 3b, when the decay parameter in spreadr was manipulated to model this mechanism the cognitive network model successfully replicated the behavioral findings. The results suggest that models of cognition need to take into account the multi-scale structure that exists among representations in memory, and how that structure can influence processing.

Cognate status modulates the comprehension of isolated reduced forms

ABSTRACT There are competing theories about the representation of reduced variants of words in the mental lexicon. At the same time, speech reduction is known to cause problems for non-natives' speech comprehension. We investigate whether processing of full and reduced forms of cognates can help to better understand how reduced forms are represented in the mental lexicon. In an English auditory lexical decision task during which the brain response (EEG) was recorded, highly proficient Dutch learners of English listened to full and reduced forms of cognates and non-cognates. In the reduced forms, schwa was omitted. This schwa reduction occurred in either poststress or prestress position. While behavioural data (accuracy and reaction time) did not yield convincing information about the status of reduced forms, EEG data strongly suggest that form representations play an important role, in both prestress and poststress words. The results have clear implications for theories and models of spoken word recognition.

COSMO-Onset: A Neurally-Inspired Computational Model of Spoken Word Recognition, Combining Top-Down Prediction and Bottom-Up Detection of Syllabic Onsets.

Recent neurocognitive models commonly consider speech perception as a hierarchy of processes, each corresponding to specific temporal scales of collective oscillatory processes in the cortex: 30–80 Hz gamma oscillations in charge of phonetic analysis, 4–9 Hz theta oscillations in charge of syllabic segmentation, 1–2 Hz delta oscillations processing prosodic/syntactic units and the 15–20 Hz beta channel possibly involved in top-down predictions. Several recent neuro-computational models thus feature theta oscillations, driven by the speech acoustic envelope, to achieve syllabic parsing before lexical access. However, it is unlikely that such syllabic parsing, performed in a purely bottom-up manner from envelope variations, would be totally efficient in all situations, especially in adverse sensory conditions. We present a new probabilistic model of spoken word recognition, called COSMO-Onset, in which syllabic parsing relies on fusion between top-down, lexical prediction of onset events and bottom-up onset detection from the acoustic envelope. We report preliminary simulations, analyzing how the model performs syllabic parsing and phone, syllable and word recognition. We show that, while purely bottom-up onset detection is sufficient for word recognition in nominal conditions, top-down prediction of syllabic onset events allows overcoming challenging adverse conditions, such as when the acoustic envelope is degraded, leading either to spurious or missing onset events in the sensory signal. This provides a proposal for a possible computational functional role of top-down, predictive processes during speech recognition, consistent with recent models of neuronal oscillatory processes.

Does signal reduction imply predictive coding in models of spoken word recognition?

Pervasive behavioral and neural evidence for predictive processing has led to claims that language processing depends upon predictive coding. Formally, predictive coding is a computational mechanism where only deviations from top-down expectations are passed between levels of representation. In many cognitive neuroscience studies, a reduction of signal for expected inputs is taken as being diagnostic of predictive coding. In the present work, we show that despite not explicitly implementing prediction, the TRACE model of speech perception exhibits this putative hallmark of predictive coding, with reductions in total lexical activation, total lexical feedback, and total phoneme activation when the input conforms to expectations. These findings may indicate that interactive activation is functionally equivalent or approximant to predictive coding or that caution is warranted in interpreting neural signal reduction as diagnostic of predictive coding.