Philip J. Monahan

Feedforward and feedback in speech perception: Revisiting analysis by synthesis

We revisit the analysis by synthesis (A×S) approach to speech recognition. In the late 1950s and 1960s, Stevens and Halle proposed a model of spoken word recognition in which candidate word representations were synthesised from brief cues in the auditory signal and analysed against the input signal in tightly linked bottom-up/top-down fashion. While this approach failed to garner much support at the time, recent years have brought a surge of interest in Bayesian approaches to perception, and the idea of A×S has consequently gained attention, particularly in the domain of visual perception. We review the model and illustrate some data from speech perception that are well-accounted for in the context of such an architecture. We focus on prediction in speech perception, an operation at the centre of the A×S algorithm. The data reviewed here and the current possibilities to study online measures of speech processing using cognitive neuroscience methods, in our view, add to a provocative series of arguments why A×S should be reconsidered as a contender in speech recognition research, complementing currently more dominant models.

Evidence for early morphological decomposition: Combining masked priming with magnetoencephalography

Are words stored as morphologically structured representations? If so, when during word recognition are morphological pieces accessed? Recent masked priming studies support models that assume early decomposition of (potentially) morphologically complex words. The electrophysiological evidence, however, is inconsistent. We combined masked morphological priming with magneto-encephalography (MEG), a technique particularly adept at indexing processes involved in lexical access. The latency of an MEG component peaking, on average, 220 msec post-onset of the target in left occipito-temporal brain regions was found to be sensitive to the morphological prime–target relationship under masked priming conditions in a visual lexical decision task. Shorter latencies for related than unrelated conditions were observed both for semantically transparent (cleaner–CLEAN) and opaque (corner–CORN) prime–target pairs, but not for prime–target pairs with only an orthographic relationship (brothel–BROTH). These effects are likely to reflect a prelexical level of processing where form-based representations of stems and affixes are represented and are in contrast to models positing no morphological structure in lexical representations. Moreover, we present data regarding the transitional probability from stem to affix in a post hoc comparison, which suggests that this factor may modulate early morphological decomposition, particularly for opaque words. The timing of a robust MEG component sensitive to the morphological relatedness of prime–target pairs can be used to further understand the neural substrates and the time course of lexical processing.

Auditory sensitivity to formant ratios: Toward an account of vowel normalisation

A long-standing question in speech perception research is how listeners extract linguistic content from a highly variable acoustic input. In the domain of vowel perception, formant ratios, or the calculation of relative bark differences between vowel formants, have been a sporadically proposed solution. We propose a novel formant ratio algorithm in which the first (F1) and second (F2) formants are compared against the third formant (F3). Results from two magnetoencephalographic experiments are presented that suggest auditory cortex is sensitive to formant ratios. Our findings also demonstrate that the perceptual system shows heightened sensitivity to formant ratios for tokens located in more crowded regions of the vowel space. Additionally, we present statistical evidence that this algorithm eliminates speaker-dependent variation based on age and gender from vowel productions. We conclude that these results present an impetus to reconsider formant ratios as a legitimate mechanistic component in the solution to the problem of speaker normalisation.

Speech Perception Cognitive Foundations and Cortical Implementation

Speech perception includes, minimally, the set of computations that transform continuously varying acoustic signals into linguistic representations that can be used for subsequent processing. The auditory and motor subroutines of this complex perceptual process are executed in a network of brain areas organized in ventral and dorsal parallel pathways, performing sound-to-meaning and sound-to-motor mappings, respectively. Research on speech using neurobiological techniques argues against narrow motor or auditory theories. To account for the range of cognitive and neural attributes, integrative computational models seem promising.

Underspecification and asymmetries in voicing perception

The purpose of our study is to show that phonological knowledge is an important basis for making predictions during speech perception. Taking the phonological constraint in English that coda obstruent clusters agree in their value for voicing, we conducted two experiments using vowel–stop–fricative sequences, where the task was to identify the fricative. Stimuli included sequences that were either congruent or incongruent. Consistent with models of featural underspecification for voiceless obstruents, our results indicate that only voiced stops induced predictions for an upcoming voiced fricative, eliciting processing difficulty when such predictions were not met. In contrast, voiceless stops appear to induce no equivalent predictions. These results demonstrate the important role of abstract phonological knowledge in online processing, and the asymmetries in our findings also suggest that only specified features are the basis for generating perceptual predictions about the upcoming speech signal.

Masked repetition priming using magnetoencephalography

Masked priming is used in psycholinguistic studies to assess questions about lexical access and representation. We present two masked priming experiments using MEG. If the MEG signal elicited by words reflects specific aspects of lexical retrieval, then one expects to identify specific neural correlates of retrieval that are sensitive to priming. To date, the electrophysiological evidence has been equivocal. We report findings from two experiments. Both employed identity priming, where the prime and target are the same lexical item but differ in case (NEWS-news). The first experiment used only forward masking, while the prime in the second experiment was both preceded and followed by a mask (backward masking). In both studies, we find a significant behavioral effect of priming. Using MEG, we identified a component peaking approximately 225 ms post-onset of the target, whose latency was sensitive to repetition. These findings support the notion that properties of the MEG response index specific lexical processes and demonstrate that masked priming can be effectively combined with MEG to investigate the nature of lexical processing.

Numbers are not like words: Different pathways for literacy and numeracy

Literacy and numeracy are two fundamental cognitive skills that require mastering culturally-invented symbolic systems for representing spoken language and quantities. How numbers and words are processed in the human brain and their temporal dynamics remain unclear. Using MEG (magnetoencephalography), we find brain activation differences for literacy and numeracy from early stages of processing in the temporal-occipital and temporal-parietal regions. Native speakers of Spanish were exposed to visually presented words, pseudowords, strings of numbers, strings of letters and strings of symbols while engaged in a go/no-go task. Results showed more evoked neuromagnetic activity for words and pseudowords compared to symbols at ~120-130ms in the left occipito-temporal and temporal-parietal cortices (angular gyrus and intra-parietal sulcus) and at ~200ms in the left inferior frontal gyrus and left temporal areas. In contrast, numbers showed more activation than symbols at similar time windows in homologous regions of the right hemisphere: occipito-temporal and superior and middle temporal cortices at ~100-130ms. A direct comparison between the responses to words and numbers confirmed this distinct lateralization for the two stimulus types. These results suggest that literacy and numeracy follow distinct processing streams through the left and right hemispheres, respectively, and that the temporal-parietal and occipito-temporal regions may interact during processing alphanumeric stimuli.

Neuromagnetic evidence for early auditory restoration of fundamental pitch.

Background Understanding the time course of how listeners reconstruct a missing fundamental component in an auditory stimulus remains elusive. We report MEG evidence that the missing fundamental component of a complex auditory stimulus is recovered in auditory cortex within 100 ms post stimulus onset. Methodology Two outside tones of four-tone complex stimuli were held constant (1200 Hz and 2400 Hz), while two inside tones were systematically modulated (between 1300 Hz and 2300 Hz), such that the restored fundamental (also knows as “virtual pitch”) changed from 100 Hz to 600 Hz. Constructing the auditory stimuli in this manner controls for a number of spectral properties known to modulate the neuromagnetic signal. The tone complex stimuli only diverged on the value of the missing fundamental component. Principal Findings We compared the M100 latencies of these tone complexes to the M100 latencies elicited by their respective pure tone (spectral pitch) counterparts. The M100 latencies for the tone complexes matched their pure sinusoid counterparts, while also replicating the M100 temporal latency response curve found in previous studies. Conclusions Our findings suggest that listeners are reconstructing the inferred pitch by roughly 100 ms after stimulus onset and are consistent with previous electrophysiological research suggesting that the inferential pitch is perceived in early auditory cortex.

The Cambridge Handbook of Biolinguistics: Computational primitives in phonology and their neural correlates

Understanding the representational and algorithmic systems that underlie language has been the primary goal for linguists (e.g., Chomsky 1959a). As it stands, however, the neural bases of linguistic systems remain woefully underspecified. Moving toward an understanding of how such systems are encoded in the brain, however, demands a linking between the basic ontological primitives that underlie linguistic systems on the one hand with neurophysiology on the other (Poeppel and Embick 2005). The goal of the current chapter is to identify what we believe to be the core computational primitives that underlie phonological knowledge and present evidence from the domain of cognitive neuroscience that attempts to investigate the nature of the neural correlates of these primitives. We take the goal of biolinguistics to be to understand the biological bases of human language with a strong emphasis on its evolutionary origins (Jenkins 2000). In practice, the focus has been placed on investigating the biological underpinnings of syntactic knowledge, e.g., the evolutionary origins of the syntactic operation MERGE (Hauser, Chomsky and Fitch 2002). Although this work has been enlightening, it is perhaps surprising that relatively less effort has been paid to the biolinguistic foundations of phonological systems (Samuels 2011) and their interface with speech perception and production. This is surprising because we believe that there are a number of reasons that biolinguistic inquiry into this domain should be more tractable. First, fewer levels of abstraction separate the fundamental representations of phonology from the basic sensory input representations. This means that knowledge about how basic auditory information is represented and processed in both humans and animals is more likely to provide important insights into how phono-logical information could be represented and processed. For this type of

Prediction of agreement and phonetic overlap shape sublexical identification

The mapping between the physical speech signal and our internal representations is rarely straightforward. When faced with uncertainty, higher-order information is used to parse the signal and because of this, the lexicon and some aspects of sentential context have been shown to modulate the identification of ambiguous phonetic segments. Here, using a phoneme identification task (i.e., participants judged whether they heard [o] or [a] at the end of an adjective in a noun–adjective sequence), we asked whether grammatical gender cues influence phonetic identification and if this influence is shaped by the phonetic properties of the agreeing elements. In three experiments, we show that phrase-level gender agreement in Spanish affects the identification of ambiguous adjective-final vowels. Moreover, this effect is strongest when the phonetic characteristics of the element triggering agreement and the phonetic form of the agreeing element are identical. Our data are consistent with models wherein listeners generate specific predictions based on the interplay of underlying morphosyntactic knowledge and surface phonetic cues.