Phonemic perception exhibits coarticulation sensitivity, phonotactic sensitivity and lexical sensitivity. Three kinds of models of speech perception are found in the literature, which embody different answers to the question of how the three kinds of sensitivity are related to each other: two-step models, one-step models and lexicalist models. In two-step models (Church, 1987), phonemes are first extracted, and phonotactic repairs are subsequently made on the obtained phoneme string; both phonemic categorization and phonotactic repair are sublexical, and coarticulation sensitivity should only affect initial (prephonotactic) phonemic categorization. In one-step models (Dehaene-Lambertz et al., 2000; Dupoux et al., 2011; Mehler et al., 1990), phonemic categorization and phonotactic repair are sublexical and simultaneous; phonotactic repairs themselves depend on coarticulation cues. Such models can be implemented in two different versions: suprasegmental matching, according to which a speech signal is matched against phonotactics-respecting suprasegmental units (such as syllables), rather than phonemes, and slot filling, according to which a speech signal is matched against phonemes as fillers for slots in phonotactics-respecting suprasegmental units. In lexicalist models (Cutler et al., 2009; McClelland & Elman, 1986), coarticulation sensitivity and/or phonotactic sensitivity reduce to lexical sensitivity. McClelland & Elman (1986) claim a lexicalist reduction of phonotactic sensitivity; Cutler et al.’s (2009) make a claim implying lexicalist reductions both of phonotactic sensitivity and of coarticulation sensitivity. This thesis attempts to distinguish among those models. Since different perceptual processes are assumed in these three models (whether sublexical units are perceived, or how many stages are involved in perceptual processing), our understanding of how speech perception works crucially depends on the relative superiority of those three kinds of models. Based on the results available in the past literature on the one hand, and on the results of perceptual experiments with Japanese listeners testing their coarticulation sensitivity in different settings on the other, this thesis argues for the superiority of the slot filling version of one-step models over the others. According to this conclusion, phonemic parsing (categorization) and phonotactic parsing (repair) are separate but parallel sublexical processes.

This dissertation explores the relationship between English phonotactics – sequential dependencies between adjacent segments – and the metrical parse, which relies on the division of words into syllables. Most current theories of syllabification operate under the assumption that the phonotactic restrictions which co-determine syllable boundaries are constrained by word edges. For example, a syllable can never begin with a consonant sequence that is not also attested as a word onset. This view of phonotactics as categorical is outdated: for several decades now, psycholinguistic research employing monosyllables has shown that phonotactic knowledge is gradient, and that this gradience is projected from the lexicon and possibly also based on differences in sonority among consonants located at word margins. This dissertation is an attempt to reconcile syllabification theory with this modern view of phonotactics. In what follows, I propose and defend a gradient metrical parsing model which assigns English syllable boundaries as a probabilistic function of the well-formedness relations that obtain between potential syllable onsets and offsets. I argue that this well-formedness is subserved by the same sources already established in the phonotactic literature: probabilistic generalizations over the word edges as well as sonority. In support of my proposal, I provide experimental evidence from five sources: (1) a pseudoword hyphenation experiment, (2) a reanalysis of a well-known, large-scale hyphenation study using real English words, (3) a forced-choice preference task employing nonwords presented as minimal stress pairs, (4) an online stress assignment experiment, and (5) a study of the speech errors committed by the participants of (4). The results of all studies converge in support of the gradient parsing model and correlate significantly with each other. Subsequent computer simulations suggest that the gradient model is preferred to the categorical alternative throughout all stages of lexical acquisition. This dissertation contains co-authored material accepted for publication.

It is well documented in the literature that bilingual speakers simultaneously activate both languages during spoken language processing (e.g., Marian & Spivey, 2003). However, parallel activation can lead to competition between the two languages (e.g., Blumenfield & Marian, 2013; Freeman, Shook, & Marian, 2016). The Unified Competition model (UCM) provides a theory as to how bilingual speakers navigate through two languages while different linguistic cues are competing (MacWhinney, 2005). The UCM proposes that cues are used to process language, based on cue validity (the product of how reliable and available a cue is), which is determined by cue strength (a measure based on conflict reliability; how reliable a cue is when it directly conflicts with others). Two likely cues bilingual speakers use while processing a novel spoken word are linguistic environment (the language being spoken around them) and phonotactic probability (the probability of the sounds making up a novel word). Applying the theory of the UCM this study sets to answer the following general question: How do Spanish/English bilingual adults assign language membership to nonwords when linguistic environment and phonotactic cues are competing? The current study consisted of twenty-two Spanish/English adults who listened to 96 nonwords that corresponded to three different groups based on phonotactic probability: Language Exclusive (the phonotactics of the nonwords designated them as either Spanish only or English only), High-Low (the nonwords had high phonotactic probability in one language and low probability in the other), and Ambiguous (the nonowords had similar phonotactic probability in both languages). The participants were tested in one of two linguistic environments (primarily English with some Spanish code-switching or primarily Spanish with some English code-switching) and partook in a two-alternative forced choice listening test (participants determined if each nonword was either Spanish or English). The language membership decision was measured via verbal response and eye-tracking using EyeLink 1000 Plus measuring eye gaze, number of fixations and switches. In general, results indicated that Spanish/English bilingual adults relied only on phonotactic probability when making language membership decisions, but not as strongly as may be suggested by the UCM. The results of this study suggest that environmental cues are not strong enough to impact spoken language processing in Spanish/English bilingual adults and that phonotactic probability is likely a more easily accessible (and therefore more commonly used) cue.

Phonotactics are the restrictions on sound sequences within a word or syllable. They are an important cue for speech segmentation and a guiding force in the creation of new words. By studying phonotactics, we stand to gain a better understanding of why languages and speakers have phonologies. Through a series of four experiments, I will present data that sharpen our theoretical and empirical perspectives of what phonotactics are and how they are acquired.The methodology is similar to that used in studies of infant perception: children are familiarized with a set of words that contain either a few or many examples of a phonotactic sequence. The participants here are four-year-olds, and the test involves producing a target phonotactic sequence in a new word. Because the test words have not been encountered before, children must generalize what they learned in the familiarization phase and apply it to their own speech. By manipulating the phonetic and phonological characteristics of the familiarization items, we can determine which factors are relevant to phonotactic learning. In these experiments, the phonetic manipulation was the number of talkers who children heard produce a familiarization word. The phonological manipulation was the number of familiarization words that shared a phonotactic pattern.The findings include instances where learning occurs and instances where it does not. First, the data show that the well-studied correlation between phonotactic probability and production accuracy in child speech can be attributed, at least partly to perceptual learning, rather than a practice effect attributable to repeated articulation. Second, the data show that perceptual learning is a process of abstraction and learning about those abstractions. It is not about making connections between stored, unelaborated exemplars because learning from the phonetic manipulation alone was insufficient for a phonotactic pattern to generalize. Furthermore, perceptual learning is not about reorganizing pre-existing symbolic knowledge, because learning from words alone is insufficient. I argue that a model which learns abstract word-forms from direct phonetic experience, then learns phonotatics from the abstract word-forms, is the most parsimonious explanation of phonotactic learning.

This study examines the role of voiceless and voiced fricatives as the first consonant in word-initial true consonant clusters and adjunct clusters. Specifically, this study sought evidence to determine whether the lack of voiced fricatives, such as /z/ and /v/, in English word-initial true and adjunct clusters is due to an active ban or an accidental gap in the language's phonotactics. This study also looked into whether the voiceless alveolar fricative /s/ is the only fricative that can play the role of adjunct segment in word-initial adjunct clusters, or whether other fricatives, such as the voiced alveolar fricative /z/, or the voiceless and voiced labiodental fricatives /f/ and /v/ could also be adjunct segments in word-initial adjunct clusters. Fourteen native English speakers were asked to pronounce a list of non-words containing word-initial clusters with /s/, /f/, /z/, and /v/ as the first consonant and /r/, /l/, /n/, /k/, and /g/ as the second consonant. The clusters were chosen to represent different voicing statuses and places of articulation for the first consonant in the cluster, in addition to differing sonority distances between the first consonant and the second consonant of the word-initial cluster. The native English speaker productions were recorded and acoustically analyzed in order to determine the exact pronunciations each speaker used for each target cluster. The results were then statistically analyzed to reveal patterns. Results showed that the lack of voiced fricatives as the first consonant in word-initial position of true clusters in English is due to an accidental gap, due to the relatively numerous correct productions of such clusters. The the lack of voiced fricatives as the first consonant in word-initial position of adjunct clusters in English, however, is due to an active ban, due to the difficulty that the native English speakers had in correctly producing such clusters. This study also concluded that while /s/ is the only adjunct segment in English, /f/ could also play that role.

Learning environment has been proposed to be a cause of age of acquisition effects in second language acquisition. Explicit learning in adults is linked to fast initial gains but poorer ultimate attainment whereas implicit learning in children requires more input but leads to greater proficiency in the long run. The current study used ERP measures to determine if explicit learning of a phonotactic pattern interferes with implicit learning of that same pattern in adults. Listeners were told to figure out the pattern of consonants that can go together in a word by listening to 16 CVCV nonsense words in which the two consonants all matched in voicing or never matched in voicing. Listeners rated novel items that fit the pattern presented in training as far more likely to fit the rule than novel items that violated that pattern, indicating that they did indeed learn the pattern. For participants who heard the voicing-mismatch language, novel items that violated the pattern elicited a larger negativity 200-400 ms after onset compared to novel items that fit the pattern. This effect was entirely distinct from what was previously observed under implicit learning conditions. Further, three patterns of data suggest that difficult explicit learning of a phonotactic pattern decreased language learning. First, differences in N400 amplitude across training blocks were reduced compared to what was observed with implicit learning. Differences in N400 amplitude in response to trained and novel items were limited to the more easily learned matched-voicing language. Second, the ERP index of implicit phonotactic learning, a larger Late Positive Component (LPC) in response to novel items that violate compared to fit the pattern, was absent under explicit learning conditions. Third, and supporting the idea that an absence of an LPC effect indicates an absence of implicit learning, the only hint of an LPC-like effect was evident for the more easily learned matched-voicing pattern. These patterns of data suggest that, at least for patterns that native speakers have exclusively implicit knowledge of, challenging explicit learning can interfere with other aspects of language learning. Adults who approach second-language acquisition with class-room style explicit learning strategies may compromise implicit learning of complex patterns that are necessary for higher levels of ultimate proficiency.

This master thesis provides a relatively detailed description of the consonant-vowel structure of standard spoken Czech. The first part covers approaches to and findings on the combinatorial system and distribution of sound units in speech; aspects of continuous speech segmentation into intonation phrases and stress groups, phonotactics of languages in general and Czech in particular, and speech rhythm are also addressed. Recordings of 12 professional speakers of Czech - comprising 6639 words and 5368 stress groups in total - have been used to create data sets that have enabled the author to describe CVCV structures of Czech words and stress groups. The results of this research present frequencies of words and stress groups and their relations to word-class dimension, frequencies of phones in words and stress groups; in this respect, both syllabic liquids and glottal stops have been taken into account. Further, the thesis includes an overview of the most frequent CVCV patterns in words and stress groups and their variability with regard to word classes. The final part focuses on how frequently consonants, vowels and their pairs and trios occur at different places within a unit. The results are continuously compared both between themselves and with the research that used the larger SYN2005 written...

The present thesis deals with rhythmic differences between Welsh English and the British Standard. It focuses on the varieties spoken in Cardiff and Aberystwyth in particular. The first part of the theoretical chapter summarises the approaches towards rhythm from the physiological, acoustic, perceptual, and phonological perspectives. The second part provides a basic description of the British Standard, Welsh, and Welsh English. It concerns itself with the existing information related to the subject matter especially as regards Welsh varieties of English. The last, third part, serves as an overview of the most common approaches towards the search of the acoustic correlates of rhythm (%V, ∆C, ∆V, PVI, varco, RR, YARD). The following chapters of the thesis present a material based study of the data obtained for the purposes of the thesis. The segmentation was carried out according to the principles proposed by Machač and Skarnitzl 2009. Rhythm was measured for four respondents for each selected location of Wales. The age span was 35-39 years for the group from Cardiff and 29-39 for that from Aberystwyth. The values measured were compared with the research of Volín and Pollák from 2009, which, among other things, provided the results of the rhythmic values for %V and ∆C for the British Standard on the...

IBM, a well-known computer giant, and Nuance, a renowned speech technology firm, have been offering numerous speech recognition applications in the recent years. The connections between these two companies and the automobile, communication, and other eight dominating industries, including banking, electronics, energy/utilities, medical/life science, insurance, media/entertainment, retail travel and transportation, are vastly expanded and flourished. Maturity of these speech technologies drives our lifestyle to a cozy level that we cannot imagine before. In April, 2011, the world class manufacturer Foxconn decided to invest 12 billion US dollars to build iPhone/iPad factories in Brazil, the largest Portuguese speaking country in the world. It is our objective to build a language system that can help us to learn Portuguese, to savor the beauty of their culture, and to widen our vision of travel and living. This thesis investigates the design and implementation strategies for a Portuguese speech recognition system. It utilizes the speech features of the 303 common Portuguese mono-syllables as the major training and recognition methodology. A training database of 10 utterances per mono-syllable is established by applying Portuguese pronunciation rules. These 10 utterances are collected through reading 5 rounds of the same mono-syllables twice with different tones. The first pronounced pattern has high pitch of tone 1, while the second one has falling pitch of tone 4. Mel-frequency cepstral coefficients, linear predicted cepstral coefficients, and hidden Markov model are used as the two syllable feature models and the recognition model respectively. Under the AMD 2.2 GHz Athlon XP 2800+ personal computer and Ubuntu 9.04 operating system environment, correct phrase recognition rates of 87.26% can be reached using phonotactical rules for a 3,900 vocabulary Portuguese phrase database. The average computation time for the Portuguese phrase system is less than 1.5 seconds, and the training time for the systems is about two hours.

Language plays an important role for understanding people, their history, culture and even technology. Many countries of the world have developed the technology of the outer space recently, and Russian is the top of the world. In 1998 Russia further launched Zarya, the first International Space Station (ISS) Module, to the outer space, and was deeply involved in the development of the ISS with the U.S.. Since the end of the World War Two, Russia has been one of the five Permanent Members in the United Nations. And then, she became one of the G8 members, an economical forum of eight industrially advanced nations. Because these informations, it is our objective to build a language system that can help us to learn Russian, to taste the beauty of her culture, and to widen our vision of technologies. This thesis investigates the design and implementation strategies for a Russian speech recognition system. It utilizes the speech features of the 514 common Russian mono-syllables as the major training and recognition methodology. The mono-syllable is established by applying Russian pronunciation rules. These 12 utterances are collected through reading 6 rounds of the same mono-syllables twice with different tones. The first pronounced pattern has high pitch of tone 1, while the second one has falling pitch of tone 4. Mel-frequency cepstral coefficients, linear predicted cepstral coefficients, and hidden Markov model are used as the two syllable feature models and the recognition model respectively. Under the AMD 2.2 GHz Athlon XP 2800+ personal computer and Ubuntu 9.04 operating system environment, correct phrase recognition rates of 86.90% and 94.83% can be reached using phonotactical rules for a 3,900 vocabulary Russian phrase database for TORFL (Test of Russian as a Foreign Language) and a 600 person name database for Russian. The average computation time for each system is less than 1.5 seconds, and the training time for the systems is about three hours.