Speech Learning Research Articles

The effects of continuous oromotor activity on speech motor learning: speech biomechanics and neurophysiologic correlates

Sustained limb motor activity has been used as a therapeutic tool for improving rehabilitation outcomes and is thought to be mediated by neuroplastic changes associated with activity-induced cortical excitability. Although prior research has reported enhancing effects of continuous chewing and swallowing activity on learning, the potential beneficial effects of sustained oromotor activity on speech improvements is not well-documented. This exploratory study was designed to examine the effects of continuous oromotor activity on subsequent speech learning. Twenty neurologically healthy young adults engaged in periods of continuous chewing and speech after which they completed a novel speech motor learning task. The motor learning task was designed to elicit improvements in accuracy and efficiency of speech performance across repetitions of eight-syllable nonwords. In addition, transcranial magnetic stimulation was used to measure the cortical silent period (cSP) of the lip motor cortex before and after the periods of continuous oromotor behaviors. All repetitions of the nonword task were recorded acoustically and kinematically using a three-dimensional motion capture system. Productions were analyzed for accuracy and duration, as well as lip movement distance and speed. A control condition estimated baseline improvement rates in speech performance. Results revealed improved speech performance following 10 min of chewing. In contrast, speech performance following 10 min of continuous speech was degraded. There was no change in the cSP as a result of either oromotor activity. The clinical implications of these findings are discussed in the context of speech rehabilitation and neuromodulation.

The vocal development of the pale spear-nosed bat is dependent on auditory feedback.

Human vocal development and speech learning require acoustic feedback, and humans who are born deaf do not acquire a normal adult speech capacity. Most other mammals display a largely innate vocal repertoire. Like humans, bats are thought to be one of the few taxa capable of vocal learning as they can acquire new vocalizations by modifying vocalizations according to auditory experiences. We investigated the effect of acoustic deafening on the vocal development of the pale spear-nosed bat. Three juvenile pale spear-nosed bats were deafened, and their vocal development was studied in comparison with an age-matched, hearing control group. The results show that during development the deafened bats increased their vocal activity, and their vocalizations were substantially altered, being much shorter, higher in pitch, and more aperiodic than the vocalizations of the control animals. The pale spear-nosed bat relies on auditory feedback for vocal development and, in the absence of auditory input, species-atypical vocalizations are acquired. This work serves as a basis for further research using the pale spear-nosed bat as a mammalian model for vocal learning, and contributes to comparative studies on hearing impairment across species.This article is part of the theme issue ‘Vocal learning in animals and humans’.

Articulatory settings and L2 English coronal consonants.

Background/aims: We explore the potential contribution of Articulatory Settings (AS) theory to L2 speech production research, testing the hypothesis that L2 segmental speech learning should involve a gradual, overall shift in both place and constriction degree, simultaneously affecting all consonants of a series as opposed to a set of parallel but unrelated changes in learners' production of individual sounds. Methods: We conducted an electropalatography study of four francophone learners' production of French and English word-initial and -medial /t d s z n l/ via carrier-sentence reading tasks. Results: L1-L2 differences in tongue shape are more common than those in constriction location, first and foremost for sonorants, and, thus, our results are not completely consistent with AS theory's claims. Conclusions: AS theory provides a potentially rich framework for exploring the L2 speech learning of consonantal phenomena including low-level L1-L2 differences in place of articulation. We propose that the observed lack of systematic between-language articulatory differences could be attributed to a number of factors to be explored in future research, such as the targeting of voicing and manner differences before the adjustment of small place differences as well as individual patterns of entrenchment of L1 articulatory routines.

The effect of a collaborative translation task on the learning and retention of pragmatic knowledge

ABSTRACT As language acquisition research supports the value of the interactive scaffolded environment and awareness-raising communicative tasks for the development of pragmatic knowledge, the present study attempted to investigate the effect of Collaborative Translation Task (CTT) on learning and retention of request speech act strategies. The main objective of the study was to determine whether collaborative translation tasks, either from or to the target language, presented in a supportive expert-novice environment, might lead to more efficient learning and retention of pragmatic knowledge. For this purpose, an experimental study was designed with four experimental groups and one control group, receiving different types of treatment including CTTs, and structured input with and without explicit instruction. Three equivalent forms of discourse completion tasks (DCT) were given to 150 low-intermediate participants as a pre-test, a post-test and a two-month follow-up delayed post-test. The results of the study indicated that CTTs appeared to have enabled noticing and deeper processing of both pragmalinguistic and sociopragmatic knowledge and resulted in improved learning and retention of pragmatic knowledge. The results also provide positive evidence for House's (2008) hypothesis that translation has the potential to raise pragmatic and cross-cultural awareness.

Exploring phonetic cues to persuasive oral presentation: A study with British English speakers and English L2 learners

Persuasiveness in oral communication in English can be expressed with various vocal phonetic cues that may not be readily accessible to English second language (L2) learners whose native language may employ a different set of cues. With a goal to increase L2 learners’ perceived spoken confidence and persuasiveness, and obtain empirical evidence for phonetic adjustments that native English speakers make to influence listeners, the current study explored the use of consonant prolongation in stressed syllable onsets for emphasis by native British English speakers and English L2 learners. The native speakers’ durations of continuant consonants and voiceless stop consonant voice onset times (VOTs) in (1) neutrally-produced speech and (2) persuasively delivered motivational/shocking/emotional messages were compared to Latvian L2 English speakers’ productions. The results revealed that in persuasive speech, the British speakers’ consonantal durations, particularly those of continuants, got significantly longer relative to the vowels that followed them; for English L2 learners, the duration of consonants did not change as a factor of speech type. This is in line with our previous research with American English speakers and carries implications for L2 speech learning and teaching.

A Perceptual Learning Approach for Dysarthria Remediation: An Updated Review.

Purpose Early studies of perceptual learning of dysarthric speech, those summarized in Borrie, McAuliffe, and Liss (2012), yielded preliminary evidence that listeners could learn to better understand the speech of a person with dysarthria, revealing a potentially promising avenue for future intelligibility interventions. Since then, a programmatic body of research grounded in models of perceptual processing has unfolded. The current review provides an updated account of the state of the evidence in this area and offers direction for moving this work toward clinical implementation. Method The studies that have investigated perceptual learning of dysarthric speech (N = 24) are summarized and synthesized first according to the proposed learning source and then by highlighting the parameters that appear to mediate learning, culminating with additional learning outcomes. Results The recent literature has established strong empirical evidence of intelligibility improvements following familiarization with dysarthric speech and a theoretical account of the mechanisms that facilitate improved processing of the neurologically degraded acoustic signal. Conclusions There are no existing intelligibility interventions for individuals with dysarthria who cannot behaviorally modify their speech. However, there is now robust support for the development of an approach that shifts the weight of behavioral change from speaker to listener, exploiting perceptual learning to ease the intelligibility burden of dysarthria. To move this work from bench to bedside, recommendations for translational studies that establish best practices and candidacy for listener-targeted dysarthria remediation, perceptual training, are provided.

Mispronunciation of High Front and Low Hausa Vowels among the Yorùbá Speakers

Pronunciation in second language learning is sometimes challenging, especially the vowels. Vowels such as [i] and [a] are found both in Hausa and Yorùbá but [i:] and [a:] are peculiar to Hausa alone. While Hausa has short and long vowels, Yorùbá has only oral and nasal vowels in their vowel inventories. Such phonemic differences constitute learning challenges, especially for the Yorùbá native speakers. This is a cross-sectional study design using mixed methods to examines the production of high front vowels: [i], and [i:], as well as low: [a], and [a:] Hausa vowels by the Yorùbá speakers to identify which group perform better between group 1 (Yorùbá native speakers who learned Hausa in the secondary school before going to the college of education), and group 2 (Yorùbá native speakers who learned Hausa informally before going to the college of education). The study also seeks to find out vowel substitutions that occur in the pronunciation tasks using 80 participants from 18 years old and above from the College of Education system in Nigeria who were selected based on purposive sampling. The findings were discussed in line with Flege & Bohn’s (2020) ‘Revised Speech Learning Model’. 8 stimuli were audio-recorded, transcribed, and rated by two independent raters, in addition to participant observation techniques adapted. The results of the Mann-Whitney test revealed that group 2 performed better than group 1. The study discovered also that the short [a] in the first and second syllables had the highest frequency of substitution compared to [i], [i:] and [a:] vowels. Such problems have pedagogical implications for learning Hausa as a second language.

Production of L2 Italian voiced palatal lateral and voiced palatal nasal by English-speaking learners

This study examines the L2 production of the Italian (e.g. ‹tovaglia›, ‘tablecloth’) and (e.g. ‹agnello›, ‘lamb’) by English-speaking learners. Four beginner English-native speakers, one advanced English-native speakers and two Italian-native speakers completed a picture-naming task, a reading task and a language background questionnaire. An auditory and an acoustic analysis were conducted, where F1, F2, F3 and F4, and duration were measured. The results showed that both sounds are difficult for second language learners to acquire in a native-like manner. Moreover, each of these complex sounds may be produced as a sequence of two existing first languages sounds. Our findings have implications for L2 models of speech learning. We propose that a (marked) L2 sound may be produced as a sequence of existing L1 sounds.

First-language-specific orthographic effects in second-language speech

We investigated first-language (L1) orthographic effects on second-language (L2) speech production in Korean–English and Farsi–English bilinguals, as compared to English monolinguals. We used a word-reading and word-naming task to compare the production of the single grapheme (letter) (e.g, ) with the digraph (e.g., ). An acoustic analysis of 600 tokens in Praat revealed that Korean–English bilinguals exhibited significantly longer [m:] productions compared to English monolinguals, but that the Farsi–English bilinguals did not. Longer/geminate [m:] productions are attributed to orthography-induced L1 transfer. We concluded that orthography does affect L2 word-reading and phonological mental representations, even when the L1 and L2 may have different scripts. We recommend that L2 speech learning be treated as a multi-modal event.

Assessing L2 English speaking using automated scoring technology: examining automarker reliability

ABSTRACT Recent advances in machine learning have made automated scoring of learner speech widespread, and yet validation research that provides support for applying automated scoring technology to assessment is still in its infancy. Both the educational measurement and language assessment communities have called for greater transparency in describing scoring algorithms and research evidence about the reliability of automated scoring. This paper reports on a study that investigated the reliability of an automarker using candidate responses produced in an online oral English test. Based on ‘limits of agreement’ and multi-faceted Rasch analyses on automarker scores and individual examiner scores, the study found that the automarker, while exhibiting excellent internal consistency, was slightly more lenient than examiner fair average scores, particularly for low-proficiency speakers. Additionally, it was found that an automarker uncertainty measure termed Language Quality, which indicates the confidence of speech recognition, was useful for predicting automarker reliability and flagging abnormal speech.

Foreign-Language Phonetic Development Leads to First-Language Phonetic Drift: Plosive Consonants in Native Portuguese Speakers Learning English as a Foreign Language in Brazil

Fifty-six Portuguese speakers born and raised in Brazil produced Portuguese words beginning in one of four plosives, /p b k ɡ/. Twenty-eight of them were monolinguals (controls), and the rest were learners of English as a foreign language (EFL). The learners were also asked to produce English words beginning with one of four plosives, /p b k ɡ/. We measured the plosives’ voice onset times (VOT) to address the following research questions: Do foreign-language learners, whose exposure to native English oral input is necessarily limited, form new sound categories specific to their additional language? Does engaging in the learning of a foreign language affect the phonetics of one’s native language? The EFL learners were found to differ from the controls in their production of Portuguese voiced (but not voiceless) plosives—prevoicing was longer in learner speech. The learners displayed different VOT targets for voiced (but not voiceless) consonants as a function of the language they were speaking—prevoicing was longer in Portuguese. In EFL learners’ productions, English sounds appear to be fundamentally modeled on phonologically similar native sounds, but some phonetic development (or reorganization) is found. Phonetic development induced by foreign-language learning may lead to a minor reconfiguration of the phonetics of native language sounds. EFL learners may find it challenging to learn the pronunciation patterns of English, likely due to the reduced access to native oral input.

Emerging native-similar neural representations underlie non-native speech category learning success.

Learning non-native phonetic categories in adulthood is an exceptionally challenging task, characterized by large inter-individual differences in learning speed and outcomes. The neurobiological mechanisms underlying the inter-individual differences in the learning efficacy are not fully understood. Here we examined the extent to which training-induced neural representations of non-native Mandarin tone categories in English listeners (n = 53) are increasingly similar to those of the native listeners (n = 33) who acquired these categories early in infancy. We particularly assessed whether the neural similarities in representational structure between non-native learners and native listeners are robust neuromarkers of inter-individual differences in learning success. Using inter-subject neural representational similarity (IS-NRS) analysis and predictive modeling on two functional magnetic resonance imaging (fMRI) datasets, we examined the neural representational mechanisms underlying speech category learning success. Learners' neural representations that were significantly similar to the native listeners emerged in brain regions mediating speech perception following training; the extent of the emerging neural similarities with native listeners significantly predicted the learning speed and outcome in learners. The predictive power of IS-NRS outperformed models with other neural representational measures. Furthermore, neural representations underlying successful learning are multidimensional but cost-efficient in nature. The degree of the emergent native-similar neural representations was closely related to the robust neural sensitivity to feedback in the frontostriatal network. These findings provide important insights on experience-dependent representational neuroplasticity underlying successful speech learning in adulthood and could be leveraged in designing individualized feedback-based training paradigms that maximize learning efficiency.

A Potential Role for Reinforcement Learning in Speech Production.

Reinforcement learning, the ability to change motor behavior based on external reward, has been suggested to play a critical role in early stages of speech motor development and is widely used in clinical rehabilitation for speech motor disorders. However, no current evidence exists that demonstrates the capability of reinforcement to drive changes in human speech behavior. Speech provides a unique test of the universality of reinforcement learning across motor domains: Speech is a complex, high-dimensional motor task whose goals do not specify a task to be performed in the environment but ultimately must be self-generated by each speaker such that they are understood by those around them. Across four experiments, we examine whether reinforcement learning alone is sufficient to drive changes in speech behavior and parametrically test two features known to affect reinforcement learning in reaching: how informative the reinforcement signal is as well as the availability of sensory feedback about the outcomes of one's motor behavior. We show that learning does occur and is more likely when participants receive auditory feedback that gives an implicit target for production, although they do not explicitly imitate that target. Contrary to results from upper limb control, masking feedback about movement outcomes has no effect on speech learning. Together, our results suggest a potential role for reinforcement learning in speech but that it likely operates differently than in other motor domains.

Contact-induced Phonological Change of the Phoneme /s/ in the Speech of EFL Learners and Teachers in Algeria: A Case Study

This paper investigates the phenomenon of contact-induced phonological change of the phoneme /s/ in the speech of EFL learners and teachers at the secondary schools and even Higher Education in Algeria. It is perceived that the letter ‘s’ is recurrently pronounced /z/ whenever it is positioned between two vowel sounds and that the double –ss- between two vowels is simply uttered /s/ in many words. This fact is attributed to the overgeneralization of French phonological rules however this is not the case in English. Such phonological aspects have become prominent symptoms within English as Foreign Language (EFL) classrooms prior to French overuse within Algeria’s speech repertoire due to historical and practical reasons. This study seems as a departure from the meeting point of looking at such phenomenon in teachers and/or learners’ natural conversations in EFL classrooms to a critical examination of its uses in numerous English terms. To attain this, Labov’s “observer’s paradox” (1972) was considered to overcome problems while collecting spontaneous speech data. Two different strategies were employed, structured and unstructured observations, throughout thirteen years. A total random sample of 257 respondents had been included, 204 of whom are secondary school pupils and university students, and the remaining are secondary school (trainee) teachers and university lecturers. The first group was intentionally asked to derive either nouns or adjectives from a range of words containing phoneme /s/, classify some words in a table according to the final ‘-d’ pronunciation or give the opposites –especially using the prefix ‘dis_’ – to certain words keeping the same root as well pronounce them at the end. The second group was recorded. Outstandingly, the study concludes that 98% of our informants mispronounce the given words. At last, some recommendations have been suggested to alleviate the problem.

JustSpeak: Automated, User-Configurable, Interactive Agents for Speech Tutoring

Conversational agents are widely used in many situations, especially for speech tutoring. However, their contents and functions are often pre-defined and not customizable for people without technical backgrounds, thus significantly limiting their flexibility and usability. Besides, conventional agents often cannot provide feedback in the middle of training sessions because they lack technical approaches to evaluate users' speech dynamically. We propose JustSpeak: automated and interactive speech tutoring agents with various configurable feedback mechanisms, using any speech recordings with its transcription text as the template for speech training. In JustSpeak, we developed an automated procedure to generate customized tutoring agents from user-inputted templates. Moreover, we created a set of methods to dynamically synchronize speech recognizers' behavior with the agent's tutoring progress, making it possible to detect various speech mistakes dynamically such as being stuck, mispronunciation, and rhythm deviations. Furthermore, we identified the design primitives in JustSpeak to create different novel feedback mechanisms, such as adaptive playback, follow-on training, and passive adaptation. They can be combined to create customized tutoring agents, which we demonstrate with an example for language learning. We believe JustSpeak can create more personalized speech learning opportunities by enabling tutoring agents that are customizable, always available, and easy-to-use.

Glottalization and linking in the L2 speech of Czech learners of Spanish, Italian and Portuguese

The use of linking or glottalization contributes to the characteristic sound pattern of a language, and the use of one in place of the other may affect a speaker’s comprehensibility and fluency in certain contexts. In this study, native speakers of Czech, a language that is associated with a frequent use of glottalization in vowel-initial word onsets, are examined in the second language (L2) context of three Romance languages that predominantly employ linking between words (Spanish, Italian and Portuguese). In total, 29 native speakers and 51 non-native learners were asked to read a short text in the respective language. The learners were divided into two groups based on their experience with the target language. A number of other factors were examined in a mixed-effects logistic regression model (segmental context, lexical stress, prosodic breaks, and the semantic status of the words). The main results show that, regardless of the target language, the more experienced (ME) learners displayed significantly lower rates of glottalization than the less experienced (LE) learners, but significantly higher rates than native speakers. The pedagogical implications of the results are discussed.

Learning to recognize while learning to speak: Self-supervision and developing a speaking motor

Traditionally, learning speech synthesis and speech recognition were investigated as two separate tasks. This separation hinders incremental development for concurrent synthesis and recognition, where partially-learned synthesis and partially-learned recognition must help each other throughout lifelong learning. This work is a paradigm shift—we treat synthesis and recognition as two intertwined aspects of a lifelong learning agent. Furthermore, in contrast to existing recognition or synthesis systems, babies do not need their mothers to directly supervise their vocal tracts at every moment during the learning. We argue that self-generated non-symbolic states/actions at fine-grained time level help such a learner as necessary temporal contexts. Here, we approach a new and challenging problem—how to enable an autonomous learning system to develop an artificial speaking motor for generating temporally-dense (e.g., frame-wise) actions on the fly without human handcrafting a set of symbolic states. The self-generated states/actions are Muscles-like, High-dimensional, Temporally-dense and Globally-smooth (MHTG), so that these states/actions are directly attended for concurrent synthesis and recognition for each time frame. Human teachers are relieved from supervising learner’s motor ends. The Candid Covariance-free Incremental (CCI) Principal Component Analysis (PCA) is applied to develop such an artificial speaking motor where PCA features drive the motor. Since each life must develop normally, each Developmental Network-2 (DN-2) reaches the same network (maximum likelihood, ML) regardless of randomly initialized weights, where ML is not just for a function approximator but rather an emergent Turing Machine. The machine-synthesized sounds are evaluated by both the neural network and humans with recognition experiments. Our experimental results showed learning-to-synthesize and learning-to-recognize-through-synthesis for phonemes. This work corresponds to a key step toward our goal to close a great gap toward fully autonomous machine learning directly from the physical world.

Missense and truncating variants in CHD5 in a dominant neurodevelopmental disorder with intellectual disability, behavioral disturbances, and epilepsy

Located in the critical 1p36 microdeletion region, the chromodomain helicase DNA-binding protein 5 (CHD5) gene encodes a subunit of the nucleosome remodeling and deacetylation (NuRD) complex required for neuronal development. Pathogenic variants in six of nine chromodomain (CHD) genes cause autosomal dominant neurodevelopmental disorders, while CHD5-related disorders are still unknown. Thanks to GeneMatcher and international collaborations, we assembled a cohort of 16 unrelated individuals harboring heterozygous CHD5 variants, all identified by exome sequencing. Twelve patients had de novo CHD5 variants, including ten missense and two splice site variants. Three familial cases had nonsense or missense variants segregating with speech delay, learning disabilities, and/or craniosynostosis. One patient carried a frameshift variant of unknown inheritance due to unavailability of the father. The most common clinical features included language deficits (81%), behavioral symptoms (69%), intellectual disability (64%), epilepsy (62%), and motor delay (56%). Epilepsy types were variable, with West syndrome observed in three patients, generalized tonic–clonic seizures in two, and other subtypes observed in one individual each. Our findings suggest that, in line with other CHD-related disorders, heterozygous CHD5 variants are associated with a variable neurodevelopmental syndrome that includes intellectual disability with speech delay, epilepsy, and behavioral problems as main features.

The Influence of the Type of Background Noise on Perceptual Learning of Speech in Noise.

ObjectivesAuditory perceptual learning studies tend to focus on the nature of the target stimuli. However, features of the background noise can also have a significant impact on the amount of benefit that participants obtain from training. This study explores whether perceptual learning of speech in background babble noise generalizes to other, real-life environmental background noises (car and rain), and if the benefits are sustained over time.DesignNormal-hearing native English speakers were randomly assigned to a training (n = 12) or control group (n = 12). Both groups completed a pre- and post-test session in which they identified Bamford-Kowal-Bench (BKB) target words in babble, car, or rain noise. The training group completed speech-in-babble noise training on three consecutive days between the pre- and post-tests. A follow up session was conducted between 8 and 18 weeks after the post-test session (training group: n = 9; control group: n = 7).ResultsParticipants who received training had significantly higher post-test word identification accuracy than control participants for all three types of noise, although benefits were greatest for the babble noise condition and weaker for the car- and rain-noise conditions. Both training and control groups maintained their pre- to post-test improvement over a period of several weeks for speech in babble noise, but returned to pre-test accuracy for speech in car and rain noise.ConclusionThe findings show that training benefits can show some generalization from speech-in-babble noise to speech in other types of environmental noise. Both groups sustained their learning over a period of several weeks for speech-in-babble noise. As the control group received equal exposure to all three noise types, the sustained learning with babble noise, but not other noises, implies that a structural feature of babble noise was conducive to the sustained improvement. These findings emphasize the importance of considering the background noise as well as the target stimuli in auditory perceptual learning studies.

Discourse Markers in Learner Speech: A Corpus Based Comparative Study

Having a vital role in organizing conversational interaction, discourse markers (DMs) are considered to be a significant part of communicative competence. From a Relevance Theoretic perspective and combining insights from corpus linguistics and second language acquisition research, this study is an attempt to explore how non-native speakers of English (NNS) use the DMs in their spoken language. Following the framework of Contrastive Interlanguage Analysis (CIA), this study is based on two specific corpora: Turkish sub-corpus of The Louvain International Database of Spoken English Interlanguage (LINDSEI-TR) for NNS speech and The Louvain Corpus of Native English Conversation (LOCNEC) for native speech. The occurrences of DMs in both corpora were determined using inferential and descriptive statistics. The findings of the study are discussed in connection with implications for language practitioners and syllabus designers in the field of language teaching.