Publications

Abstract

Large auditory-visual (AV) speech corpora are the grist of modern research in speech science, but no such corpus exists for Australian English. This is unfortunate, for speech science is the brains behind speech technology and applications such as text-to-speech (TTS) synthesis, automatic speech recognition (ASR), speaker recognition and forensic identification, talking heads, and hearing prostheses. Advances in these research areas in Australia require a large corpus of Australian English. Here the authors describe a blueprint for building the Big Australian Speech Corpus (the Big ASC), a corpus of over 1,100 speakers from urban and rural Australia, including speakers of non-indigenous, indigenous, ethnocultural, and disordered forms of Australian English, each of whom would be sampled on three occasions in a range of speech tasks designed by the researchers who would be using the corpus.

Abstract

English listeners largely disregard suprasegmental cues to stress in recognizing words. Evidence for this includes the demonstration of Fear et al. [J. Acoust. Soc. Am. 97, 1893–1904 (1995)] that cross-splicings are tolerated between stressed and unstressed full vowels (e.g., au- of autumn, automata). Dutch listeners, however, do exploit suprasegmental stress cues in recognizing native-language words. In this study, Dutch listeners were presented with English materials from the study of Fear et al. Acceptability ratings by these listeners revealed sensitivity to suprasegmental mismatch, in particular, in replacements of unstressed full vowels by higher-stressed vowels, thus evincing greater sensitivity to prosodic goodness than had been shown by the original native listener group.

Abstract

Two phoneme goodness rating experiments addressed the role of orthographic knowledge in the evaluation of speech sounds. Ratings for the best tokens of /s/ were higher in words spelled with S (e.g., bless) than in words where /s/ was spelled with C (e.g., voice). This difference did not appear for analogous nonwords for which every lexical neighbour had either S or C spelling (pless, floice). Models of phonemic processing incorporating obligatory influence of lexical information in phonemic processing cannot explain this dissociation; the data are consistent with models in which phonemic decisions are not subject to necessary top-down lexical influence.

Abstract

Three experiments, in which Japanese listeners detected Japanese words embedded in nonsense sequences, examined the perceptual consequences of vowel devoicing in that language. Since vowelless sequences disrupt speech segmentation [Norris et al. (1997). Cognit. Psychol. 34, 191– 243], devoicing is potentially problematic for perception. Words in initial position in nonsense sequences were detected more easily when followed by a sequence containing a vowel than by a vowelless segment (with or without further context), and vowelless segments that were potential devoicing environments were no easier than those not allowing devoicing. Thus asa, “morning,” was easier in asau or asazu than in all of asap, asapdo, asaf, or asafte, despite the fact that the /f/ in the latter two is a possible realization of fu, with devoiced [u]. Japanese listeners thus do not treat devoicing contexts as if they always contain vowels. Words in final position in nonsense sequences, however, produced a different pattern: here, preceding vowelless contexts allowing devoicing impeded word detection less strongly (so, sake was detected less accurately, but not less rapidly, in nyaksake—possibly arising from nyakusake—than in nyagusake). This is consistent with listeners treating consonant sequences as potential realizations of parts of existing lexical candidates wherever possible.

Abstract

Recognizing word boundaries in continuous speech requires detailed knowledge of the native language. In the first year of life, infants acquire considerable word segmentation abilities. Infants at this early stage in word segmentation rely to a large extent on the metrical pattern of their native language, at least in stress-based languages. In Dutch and English (both languages with a preferred trochaic stress pattern), segmentation of strong-weak words develops rapidly between 7 and 10 months of age. Nevertheless, trochaic languages contain not only strong-weak words but also words with a weak-strong stress pattern. In this article, we present electrophysiological evidence of the beginnings of weak-strong word segmentation in Dutch 10-month-olds. At this age, the ability to combine different cues for efficient word segmentation does not yet seem to be completely developed. We provide evidence that Dutch infants still largely rely on strong syllables, even for the segmentation of weak-strong words.

Abstract

Two artificial-language learning experiments directly compared English, French, and Dutch listeners’ use of suprasegmental cues for continuous-speech segmentation. In both experiments, listeners heard unbroken sequences of consonant-vowel syllables, composed of recurring three- and four-syllable “words.” These words were demarcated by(a) no cue other than transitional probabilities induced by their recurrence, (b) a consistent left-edge cue, or (c) a consistent right-edge cue. Experiment 1 examined a vowel lengthening cue. All three listener groups benefited from this cue in right-edge position; none benefited from it in left-edge position. Experiment 2 examined a pitch-movement cue. English listeners used this cue in left-edge position, French listeners used it in right-edge position, and Dutch listeners used it in both positions. These findings are interpreted as evidence of both language-universal and language-specific effects. Final lengthening is a language-universal effect expressing a more general (non-linguistic) mechanism. Pitch movement expresses prominence which has characteristically different placements across languages: typically at right edges in French, but at left edges in English and Dutch. Finally, stress realization in English versus Dutch encourages greater attention to suprasegmental variation by Dutch than by English listeners, allowing Dutch listeners to benefit from an informative pitch-movement cue even in an uncharacteristic position.

Abstract

Languages differ in the constitution of their phonemic repertoire and in the relative distinctiveness of phonemes within the repertoire. In the present study, we asked whether such differences constrain spoken-word recognition, via two word reconstruction experiments, in which listeners turned non-words into real words by changing single sounds. The experiments were carried out in Dutch (which has a relatively balanced vowel-consonant ratio and many similar vowels) and in Spanish (which has many more consonants than vowels and high distinctiveness among the vowels). Both Dutch and Spanish listeners responded significantly faster and more accurately when required to change vowels as opposed to consonants; when allowed to change any phoneme, they more often altered vowels than consonants. Vowel information thus appears to constrain lexical selection less tightly (allow more potential candidates) than does consonant information, independent of language-specific phoneme repertoire and of relative distinctiveness of vowels.

Abstract

Spraak is continu, er zijn geen betrouwbare signalen waardoor de luisteraar weet waar het ene woord eindigt en het volgende begint. Voor volwassen luisteraars is het segmenteren van gesproken taal in afzonderlijke woorden dus niet onproblematisch, maar voor een kind dat nog geen woordenschat bezit, vormt de continuïteit van spraak een nog grotere uitdaging. Desalniettemin produceren de meeste kinderen hun eerste herkenbare woorden rond het begin van het tweede levensjaar. Aan deze vroege spraakproducties gaat een formidabele perceptuele prestatie vooraf. Tijdens het eerste levensjaar - met name gedurende de tweede helft - ontwikkelt de spraakperceptie zich van een algemeen fonetisch discriminatievermogen tot een selectieve gevoeligheid voor de fonologische contrasten die in de moedertaal voorkomen. Recent onderzoek heeft verder aangetoond dat kinderen, lang voordat ze ook maar een enkel woord kunnen zeggen, in staat zijn woorden die kenmerkend zijn voor hun moedertaal te onderscheiden van woorden die dat niet zijn. Bovendien kunnen ze woorden die eerst in isolatie werden aangeboden herkennen in een continue spraakcontext. Het dagelijkse taalaanbod aan een kind van deze leeftijd maakt het in zekere zin niet gemakkelijk, bijvoorbeeld doordat de meeste woorden niet in isolatie voorkomen. Toch wordt het kind ook wel houvast geboden, onder andere doordat het woordgebruik beperkt is.

Abstract

Dutch listeners were slower to make judgements about the semantic relatedness between a spoken target word (e.g. atLEET, 'athlete') and a previously presented visual prime word (e.g. SPORT 'sport') when the spoken word was mis-stressed. The adverse effect of mis-stressing confirms the role of stress information in lexical recognition in Dutch. However, although the erroneous stress pattern was always initially compatible with a competing word (e.g. ATlas, 'atlas'), mis-stressed words did not produced high false alarm rates in unrelated pairs (e.g. SPORT - atLAS). This suggests that stress information did not completely rule out segmentally matching but suprasegmentally mismatching words, a finding consistent with spoken-word recognition models involving multiple activation and inter-word competition.

Abstract

Simulations explored the inability of the TRACE model of spoken-word recognition to model the effects on human listening of acoustic-phonetic mismatches in word forms. The source of TRACE's failure lay not in its interactive connectivity, not in the presence of interword competition, and not in the use of phonemic representations, but in the need for continuously optimised interpretation of the input. When an analogue of TRACE was allowed to cycle to asymptote on every slice of input, an acceptable simulation of the subcategorical mismatch data was achieved. Even then, however, the simulation was not as close as that produced by the Merge model.

Abstract

Dutch-learning and English-learning 9-month-olds were tested, using the Headturn Preference Procedure, for their ability to segment Dutch words with strong/weak stress patterns from fluent Dutch speech. This prosodic pattern is highly typical for words of both languages. The infants were familiarized with pairs of words and then tested on four passages, two that included the familiarized words and two that did not. Both the Dutch- and the English-learning infants gave evidence of segmenting the targets from the passages, to an equivalent degree. Thus, English-learning infants are able to extract words from fluent speech in a language that is phonetically different from English. We discuss the possibility that this cross-language segmentation ability is aided by the similarity of the typical rhythmic structure of Dutch and English words.

Abstract

Lexical knowledge influences how human listeners make decisions about speech sounds. Positive lexical effects (faster responses to target sounds in words than in nonwords) are robust across several laboratory tasks, while negative effects (slower responses to targets in more word-like nonwords than in less word-like nonwords) have been found in phonetic decision tasks but not phoneme monitoring tasks. The present experiments tested whether negative lexical effects are therefore a task-specific consequence of the forced choice required in phonetic decision. We compared phoneme monitoring and phonetic decision performance using the same Dutch materials in each task. In both experiments there were positive lexical effects, but no negative lexical effects. We observe that in all studies showing negative lexical effects, the materials were made by cross-splicing, which meant that they contained perceptual evidence supporting the lexically-consistent phonemes. Lexical knowledge seems to influence phonemic decision-making only when there is evidence for the lexically-consistent phoneme in the speech signal.

Abstract

We briefly describe the Merge model of phonemic decision-making, and, in the light of general arguments about the possible role of feedback in spoken-word recognition, defend Merge's feedforward structure. Merge not only accounts adequately for the data, without invoking feedback connections, but does so in a parsimonious manner.

Abstract

The central thesis of the target article was that feedback is never necessary in spoken word recognition. The commentaries present no new data and no new theoretical arguments which lead us to revise this position. In this response we begin by clarifying some terminological issues which have lead to a number of significant misunderstandings. We provide some new arguments to support our case that the feedforward model Merge is indeed more parsimonious than the interactive alternatives, and that it provides a more convincing account of the data than alternative models. Finally, we extend the arguments to deal with new issues raised by the commentators such as infant speech perception and neural architecture.

Abstract

Top-down feedback does not benefit speech recognition; on the contrary, it can hinder it. No experimental data imply that feedback loops are required for speech recognition. Feedback is accordingly unnecessary and spoken word recognition is modular. To defend this thesis, we analyse lexical involvement in phonemic decision making. TRACE (McClelland & Elman 1986), a model with feedback from the lexicon to prelexical processes, is unable to account for all the available data on phonemic decision making. The modular Race model (Cutler & Norris 1979) is likewise challenged by some recent results, however. We therefore present a new modular model of phonemic decision making, the Merge model. In Merge, information flows from prelexical processes to the lexicon without feedback. Because phonemic decisions are based on the merging of prelexical and lexical information, Merge correctly predicts lexical involvement in phonemic decisions in both words and nonwords. Computer simulations show how Merge is able to account for the data through a process of competition between lexical hypotheses. We discuss the issue of feedback in other areas of language processing and conclude that modular models are particularly well suited to the problems and constraints of speech recognition.

Abstract

Two word-spotting experiments are reported that examine whether the Possible-Word Constraint (PWC) [1] is a language-specific or language-universal strategy for the segmentation of continuous speech. The PWC disfavours parses which leave an impossible residue between the end of a candidate word and a known boundary. The experiments examined cases where the residue was either a CV syllable with a lax vowel, or a CVC syllable with a schwa. Although neither syllable context is a possible word in English, word-spotting in both contexts was easier than with a context consisting of a single consonant. The PWC appears to be language-universal rather than language-specific.

Abstract

Simulations explored the inability of the TRACE model of spoken-word recognition to model the effects on human listening of subcategorical mismatch in word forms. The source of TRACE's failure lay not in interactive connectivity, not in the presence of inter-word competition, and not in the use of phonemic representations, but in the need for continuously optimised interpretation of the input. When an analogue of TRACE was allowed to cycle to asymptote on every slice of input, an acceptable simulation of the subcategorical mismatch data was achieved. Even then, however, the simulation was not as close as that produced by the Merge model, which has inter-word competition, phonemic representations and continuous optimisation (but no interactive connectivity).

Abstract

A database is presented of relative familiarity ratings for 24 sets of Japanese words, each set comprising words overlapping in the initial portions. These ratings are useful for the generation of material sets for research in the recognition of spoken words.

Abstract

Because stress can occur in any position within an Eglish word, lexical prosody could serve as a minimal distinguishing feature between pairs of words. However, most pairs of English words with stress pattern opposition also differ vocalically: OBject an obJECT, CONtent and content have different vowels in their first syllables an well as different stress patters. To test whether prosodic information is made use in auditory word recognition independently of segmental phonetic information, it is necessary to examine pairs like FORbear – forBEAR of TRUSty – trusTEE, semantically unrelated words which echbit stress pattern opposition but no segmental difference. In a cross-modal priming task, such words produce the priming effects characteristic of homophones, indicating that lexical prosody is not used in the same was as segmental structure to constrain lexical access.

Abstract

Two bodies of recent research from experimental psycholinguistics are summarised, each of which is centred upon a concept from phonology: LEXICAL STRESS and the SYLLABLE. The evidence indicates that neither construct plays a role in prelexical representations during speech recog- nition. Both constructs, however, are well supported by other performance evidence. Testing phonological claims against performance evidence from psycholinguistics can be difficult, since the results of studies designed to test processing models are often of limited relevance to phonological theory.

Abstract

Four studies are reported in which young children’s response time to detect word targets was measured. Children under about six years of age did not show response time advantage for accented target words which adult listeners show. When semantic focus of the target word was manipulated independently of accent, children of about five years of age showed an adult-like response time advantage for focussed targets, but children younger than five did not. Id is argued that the processing advantage for accented words reflect the semantic role of accent as an expression of sentence focus. Processing advantages for accented words depend on the prior development of representations of sentence semantic structure, including the concept of focus. The previous literature on the development of prosodic competence shows an apparent anomaly in that young children’s productive skills appear to outstrip their receptive skills; however, this anomaly disappears if very young children’s prosody is assumed to be produced without an underlying representation of the relationship between prosody and semantics.

Abstract

Speech segmentation procedures may differ in speakers of different languages. Earlier work based on French speakers listening to French words suggested that the syllable functions as a segmentation unit in speech processing. However, while French has relatively regular and clearly bounded syllables, other languages, such as English, do not. No trace of syllabifying segmentation was found in English listeners listening to English words, French words, or nonsense words. French listeners, however, showed evidence of syllabification even when they were listening to English words. We conclude that alternative segmentation routines are available to the human language processor. In some cases speech segmentation may involve the operation of more than one procedure

Abstract

Infants acquire whatever language is spoken in the environment into which they are born. The mental capability of the newborn child is not biased in any way towards the acquisition of one human language rather than another. Because psychologists who attempt to model the process of language comprehension are interested in the structure of the human mind, rather than in the properties of individual languages, strategies which they incorporate in their models are presumed to be universal, not language-specific. In other words, strategies of comprehension are presumed to be characteristic of the human language processing system, rather than, say, the French, English, or Igbo language processing systems. We report here, however, on a comprehension strategy which appears to be used by native speakers of French but not by native speakers of English.

Abstract

Spontaneous self-corrections in speech pose a communication problem; the speaker must make clear to the listener not only that the original Utterance was faulty, but where it was faulty and how the fault is to be corrected. Prosodic marking of corrections - making the prosody of the repair noticeably different from that of the original utterance - offers a resource which the speaker can exploit to provide the listener with such information. A corpus of more than 400 spontaneous speech repairs was analysed, and the prosodic characteristics compared with the syntactic and semantic characteristics of each repair. Prosodic marking showed no relationship at all with the syntactic characteristics of repairs. Instead, marking was associated with certain semantic factors: repairs were marked when the original utterance had been actually erroneous, rather than simply less appropriate than the repair; and repairs tended to be marked more often when the set of items encompassing the error and the repair was small rather than when it was large. These findings lend further weight to the characterization of accent as essentially semantic in function.

Abstract

If a conversation is to proceed smoothly, the participants have to take turns to speak. Studies of conversation have shown that there are signals which speakers give to inform listeners that they are willing to hand over the conversational turn1−4. Some of these signals are part of the text (for example, completion of syntactic segments), some are non-verbal (such as completion of a gesture), but most are carried by the pitch, timing and intensity pattern of the speech; for example, both pitch and loudness tend to drop particularly low at the end of a speaker's turn. When one speaker interrupts another, the two can be said to be disputing who has the turn. Interruptions can occur because one participant tries to dominate or disrupt the conversation. But it could also be the case that mistakes occur in the way these subtle turn-yielding signals are transmitted and received. We demonstrate here that many interruptions in an interview with Mrs Margaret Thatcher, the British Prime Minister, occur at points where independent judges agree that her turn appears to have finished. It is suggested that she is unconsciously displaying turn-yielding cues at certain inappropriate points. The turn-yielding cues responsible are identified.