Phrase Boundaries Research Articles

The Multidimensional Battery of Prosody Perception (MBOPP).

Prosody can be defined as the rhythm and intonation patterns spanning words, phrases and sentences. Accurate perception of prosody is an important component of many aspects of language processing, such as parsing grammatical structures, recognizing words, and determining where emphasis may be placed. Prosody perception is important for language acquisition and can be impaired in language-related developmental disorders. However, existing assessments of prosodic perception suffer from some shortcomings. These include being unsuitable for use with typically developing adults due to ceiling effects and failing to allow the investigator to distinguish the unique contributions of individual acoustic features such as pitch and temporal cues. Here we present the Multi-Dimensional Battery of Prosody Perception (MBOPP), a novel tool for the assessment of prosody perception. It consists of two subtests: Linguistic Focus, which measures the ability to hear emphasis or sentential stress, and Phrase Boundaries, which measures the ability to hear where in a compound sentence one phrase ends, and another begins. Perception of individual acoustic dimensions (Pitch and Duration) can be examined separately, and test difficulty can be precisely calibrated by the experimenter because stimuli were created using a continuous voice morph space. We present validation analyses from a sample of 59 individuals and discuss how the battery might be deployed to examine perception of prosody in various populations.

Pitch accents and boundary tones in narrow focus statements in Basque Spanish

This presentation examines the intonation of non-contrastive narrow focus in 15 Spanish-Basque bilinguals from the Basque Country, Spain. A total of 150 semi-spontaneous statements with sentence-final new-information focus were analyzed in Praat following Spanish ToBi conventions (Aguilar et al., 2009). We report the pitch contour of prenuclear and nuclear accents and the prevalence of intermediate phrase boundaries (ip) before the focal element, none of which have been previously investigated in this Spanish variety. Results show that while the first prenuclear accent is generally delayed regardless of bilingualism (93%), the second one is delayed more often in Basque-dominant speakers. Rising-falling nuclear contours (L + H* L%), reported for broad focus in Basque Spanish (Robles Puente, 2012) and for narrow focus in other Spanish dialects (Aguilar et al., 2009), were most common in Spanish-dominant participants, while Basque-dominant participants tended to show falling contours (L* L%). Pre-focal ip boundaries occurred in 69% of sentences, regardless of bilingualism degree. Ip boundaries were typically rising, mostly involving continuation rises, although high plateaus were also attested. We will connect these novel findings to the intonational characteristics of narrow focus in Bizkaian Basque (Elordieta & Hualde, 2014) and other Peninsular Spanish dialects (Face, 2000; 2008).

Korean Prosody Phrase Boundary Prediction Model for Speech Synthesis Service in Smart Healthcare

Speech processing technology has great potential in the medical field to provide beneficial solutions for both patients and doctors. Speech interfaces, represented by speech synthesis and speech recognition, can be used to transcribe medical documents, control medical devices, correct speech and hearing impairments, and assist the visually impaired. However, it is essential to predict prosody phrase boundaries for accurate natural speech synthesis. This study proposes a method to build a reliable learning corpus to train prosody boundary prediction models based on deep learning. In addition, we offer a way to generate a rule-based model that can predict the prosody boundary from the constructed corpus and use the result to train a deep learning-based model. As a result, we have built a coherent corpus, even though many workers have participated in its development. The estimated pairwise agreement of corpus annotations is between 0.7477 and 0.7916 and kappa coefficient (K) between 0.7057 and 0.7569. In addition, the deep learning-based model based on the rules obtained from the corpus showed a prediction accuracy of 78.57% for the three-level prosody phrase boundary, 87.33% for the two-level prosody phrase boundary.

An exploration into Penultimate and Final Lengthening in Tswana (Southern Bantu)

This study investigates the segmental lengthening patterns resulting from prosodic boundaries in Tswana, a Southern Bantu language. The aim is to shed light on the interaction between Penultimate Lengthening and Final Lengthening, providing the first quantitative investigation of these phenomena in Tswana. We conducted a production experiment that applies a widely tested design to elicit production data of two different phrasal structures in coordinated noun phrases. The results suggest that Penultimate Lengthening and Final Lengthening constitute independent mechanisms, which both apply in Tswana. Penultimate Lengthening occurs before prosodic phrase boundaries as well as before word boundaries, yet at differing degrees. Before phrase boundaries, it involves a strong lengthening effect on the vowel of the penultimate syllable. Before word boundaries, the amount of lengthening is smaller. Final lengthening operates on the final syllable before a phrase boundary, involving a larger amount on the final vowel than on the preceding consonant. This pattern is in line with the pattern observed in other languages. The amount of lengthening on the final vowel is comparable to the amount on the penultimate vowel. Given that a large increase of lengthening on the penultimate syllable has not been observed in connection with Final Lengthening, we assume that Penultimate Lengthening constitutes a language-specific mechanism that applies independently. Final Lengthening, on the other hand, might be a universal phenomenon. The perceptual salience of Penultimate Lengthening, which has been widely reported in the literature for Bantu languages, might have to do with the dynamics within the lengthening domains, namely that the lengthening in penultimate position is abrupt and relatively stronger than in final position when compared to the preceding syllable.

Prosodic prominence in a stressless language: An acoustic investigation of Indonesian

Although it has been proposed that all languages may have some lexical stress property, recent studies of (Standard) Indonesian have concluded, based primarily on perception, that lexical stress is not present in this language. While it is philosophically problematic to prove the non-existence of a phenomenon, we examine data from a large-scale production study for both direct and indirect evidence of stress, contributing to the growing body of literature in this field. In the first case, evidence is sought that indicates that a particular syllable in a word exhibits acoustic properties typically associated with prominence (i.e. fundamental frequency (f0), duration, intensity, vowel quality). In the second case, evidence of enhancement of these properties on a particular syllable under focus is sought, for a more abstract stress property that is not overtly manifested at the word level. Although we find no evidence of lexical prominence, we observe acoustic patterns consistent with a higher level prominence corresponding to focus, manifested by strong (Intonational Phrase) boundary properties. Overall, our findings reveal that there is strong support for a class of languages lacking lexical stress, and in the absence of a stressed syllable to enhance, focus may be manifested prosodically as boundary properties.

Automatic detection of prosodic boundaries in spontaneous speech.

Automatic speech recognition (ASR) and natural language processing (NLP) are expected to benefit from an effective, simple, and reliable method to automatically parse conversational speech. The ability to parse conversational speech depends crucially on the ability to identify boundaries between prosodic phrases. This is done naturally by the human ear, yet has proved surprisingly difficult to achieve reliably and simply in an automatic manner. Efforts to date have focused on detecting phrase boundaries using a variety of linguistic and acoustic cues. We propose a method which does not require model training and utilizes two prosodic cues that are based on ASR output. Boundaries are identified using discontinuities in speech rate (pre-boundary lengthening and phrase-initial acceleration) and silent pauses. The resulting phrases preserve syntactic validity, exhibit pitch reset, and compare well with manual tagging of prosodic boundaries. Collectively, our findings support the notion of prosodic phrases that represent coherent patterns across textual and acoustic parameters.

Exploring the Complexity of the L2 Intonation System: An Acoustic and Eye-Tracking Study

Phonological research has demonstrated that English intonation, variably referred to as prosody, is a multidimensional and multilayered system situated at the interface of information structure, morphosyntactic structure, phonological phenomena, and pragmatic functions. The structural and functional complexity of the intonational system, however, is largely under-addressed in L2 pronunciation teaching, leading to a lack of spontaneous use of intonation despite successful imitation in classrooms. Focusing on contrastive and implicational sentence stress, this study explored the complexity of the English intonation system by investigating how L1 English and Mandarin-English L2 speakers use multiple acoustic features (i.e., pitch range, pitch level, duration, and intensity) in signaling contrastive and implicational information and how one acoustic feature (maximum pitch level) is affected by information structure (contrast), morphosyntactic structure (phrasal boundary), and a phonological phenomenon (declination) in L1 English and Mandarin-English L2 speakers' speech. Using eye-tracking technology, we also investigated (1) L1 English and Mandarin-English L2 speakers' real-time processing of lexical items that carry information structure (i.e., contrast) and typically receive stress in L1 speakers' speech; (2) the influence of visual enhancement (italics and bold) on L1 English and Mandarin-English L2 speakers' processing of contrastive information; and (3) L1 English and Mandarin-English L2 speakers' processing of pictures with contrastive information. Statistical analysis using linear mixed-effects models showed that L1 English speakers and Mandarin-English L2 speakers differed in their use of acoustic cues in signaling contrastive and implicational information. They also differed in the use of maximum pitch level in signaling sentence stress influenced by contrast, phrasal boundary, and declination. We did not find differences in L1 English and Mandarin-English L2 speakers' processing of contrastive and implicational information at the sentence level, but the two groups of participants differ in their processing of contrastive information in passages and pictures. These results suggest that processing limitations may be the reason why L2 speakers did not use English intonation spontaneously. The findings of this study also suggest that Complexity Theory (CT), which emphasizes the complex and dynamic nature of intonation, is a theoretical framework that has the potential of bridging the gap between L2 phonology and L2 pronunciation teaching.

The Syntax of Multiple Determination in Arabic: An anti- residual relative clause/close-apposition account

This paper aims to give an account of the multiple determination (determiner spreading) phenomenon in Arabic. Determiner spreading is the syntactic representation and phonological realization of multiple determiners within the same determiner phrase. As a cross-linguistic phenomenon, determiner spreading has been investigated in other languages (e.g., Scandinavian and Greek); different accounts have been proposed. For Scandinavian languages, determiner spreading has been analyzed as a representation of different semantic interpretations. As far as Greek is concerned, some analyses have been proposed; however, two prominent ones have received considerable attention in the literature: (i) a residue of a reduced relative clause and (ii) an instantiation of close appositions. Contrary to those analyses, this paper claims that none of the two analyses is suitable for Arabic; thus, a language-specific analysis is required. To analyze determiner spreading in Arabic, the current paper posits the following research question: What is the linguistic purpose of the multiple determiners found in Arabic determiner phrases? Answering the research question, the paper claims that, in addition to its indispensable role in establishing agreement between nouns and adjectives within the Arabic determiner phrase, determiner spreading demarcates syntactic and semantic phrase boundaries. The paper takes Minimalist Program and Distributed Morphology as a theoretical framework to argue that attributive adjectives are projection of an agreement phrase headed by the definite article ʔal or by the indefinite phonological marker `nunation: -n’. This proposal requires no syntactic movements in the syntax proper. The ultimate linear order is achieved in the phonological components.

A Study on the Relation between Prosodic Phrasing and Syntactic Complexity

The purpose of the current study was to examine the relation between prosodic phrasing and syntactic complexity to better understand where speakers, especially L2 learners put prosodic phrase boundaries and pauses and thus to deepen the understanding of language planning and processing. To address these questions, this study carried out a study of phonetics-syntax interface focusing on IP and ip boundaries in L2 utterances. In addition, this study was interested in learning effects of the listen-and-repeat practice in relation to syntactic complexity and readability. This study found that in L2 production, syntactic complexity and readability had significant influences on realization of prosodic phrasing and pauses, which resulted in affecting the fluency. In addition, the pronunciation practice had a positive effect on prosodic realization so that L2 learners became less susceptible to syntactic complexity and readability.

“The tiger is hitting! the duck too!” 3-year-olds can use prosodic information to constrain their interpretation of ellipsis

This work aims to investigate French children's ability to use phrasal boundaries for disambiguation of a type of ambiguity not yet studied, namely stripping sentences versus simple transitive sentences. We used stripping sentences such as “[Le tigre tape]! [Le canard aussi]!” (“[The tiger is hitting]! [The duck too]!”, in which both the tiger and the duck are hitting), which, without the prosodic information, would be ambiguous with a transitive sentence such as “[Le tigre] [tape le canard aussi]!” (“[The tiger] [is hitting the duck too]!”, in which the tiger is hitting the duck). We presented 3-to-4-year-olds and 28-month-olds with one of the two types of sentence above, while they watched two videos side-by-side on a screen: one depicting the transitive interpretation of the sentences, and another depicting the stripping interpretation. The stripping interpretation video showed the two characters as agents of the named action (e.g. a duck and a tiger hitting a bunny), and the transitive interpretation video showed only the first character as an agent, and the second character as a patient of the action (e.g. the tiger hitting the duck and the bunny). The results showed that 3-to-4-year-olds use prosodic information to correctly distinguish stripping sentences from transitive sentences, as they looked significantly more at the appropriate video, while 28-month-olds show only a trend in the same direction. While recent studies demonstrated that from 18 months of age, infants are able to use phrasal prosody to guide the syntactic analysis of ambiguous sentences, our results show that only 3-to-4-year-olds were able to reliably use phrasal prosody to constrain the parsing of stripping sentences. We discuss several factors that can explain this delay, such as differences in the frequency of these structures in child-directed speech, as well as in the complexity of the sentences and of the experimental task. Our findings add to the growing body of evidence on the role of prosody in constraining parsing in young children.

Prosodic event detection in children’s read speech

Prosody is the supra-segmental aspect of speech that helps to convey the structure and intended meaning of lexical content unambiguously. The automatic detection of prosodic events, such as phrase boundary and word prominence, has a number of applications in discourse analysis, where a combination of syntactic and acoustic-prosodic features is typically employed. This work addresses prosodic event detection in the context of assessing oral reading skills of middle-school children. We discuss the observed characteristics of a specially created labeled data set of oral reading recordings of English stories by non-native speakers. The obtained diversity of language skills adds to the known challenges of high speaker variability in the acoustic realization of prosodic events. A combination of knowledge- and data-driven feature selection is implemented to identify a compact set of word-level features from the acoustic correlates of prosody considering different ways of incorporating the necessary temporal context. The system is benchmarked with reference to a widely known prosodic event recognition system in a speaker-independent set-up to obtain a competitive performance with greatly reduced feature dimensionality. The interpretable features enable us to use the predictor model importance scores to identify high-level speaker traits that influence the acoustic realization of prosodic events, suggesting a potential extension to systems that can extract and utilize speaker idiosyncrasies for superior prosodic event detection.

Prosody-syntax interface and prosodic phrasing in brazilian portuguese

This paper focuses on the Brazilian Portuguese (PB) prosodic phrasing and has two main goals: (1) to find a correlation between the prosodic constituents boundaries, as described by the Prosodic Hierarchy Theory (Nespor & Vogel, 2007 [1986], a.o.) and the perception and production of spontaneous and reading speech breaks, and (2) to describe the phonological characteristics and the syntactic ranking of perceived and non perceived edges. The corpus under analysis includes 5 extracts both of spontaneous and reading speech lasting about 2 minutes each. The reading speech (LE) emerged from the spontaneous speech (FE) orthographic transcription which was collected from an interview in an informal environment. In the perception test, 11 referees heard the 10 speaking extracts, without punctuation, and marked the perceived breaks in the orthographic transcription of each of them. Both the 5 speakers and the 11 referees were students at UFRJ, born in Rio de Janeiro, and were between 22 and 38 years old.The results point out that the prosodic breaks are mainly perceived at the intonational phrase (I) boundary, regardless of the speech style (FE: 91%; LE 99%). However, in LE, 64% of the foreseen I boundaries, described by the Prosodic Hierarchy Theory, were perceived as breaks, but in FE, just 37% were perceived. The most usual nuclear contour in both styles is H+L* L% (this being the Portuguese neutral declarative contour), but its occurrence frequency at perceived breaks draws a distinction between LE and FE (67% and 30%, respectively). In FE, contours like L+H* H% and L*+H H% are also produced (34%). In general, descendant nuclei in LE are predominant, as well as the edge tone L; in FE, both the descendant and ascendant nuclei distribution and low or high boundaries are similar. After running a statistic test, the appearance of an L edge, as a predictive for perception, was globally significant. Concerning the syntactic boundary, it was statistically checked and the result points out that breaks are mainly perceived at the matrix phrase limit (LE: 59%; FE: 61%,), showing the endurance of the matrix phrase edge/I boundary mapping. In general, FE has proved to have a bigger variation on the relation of predicted, perceived and produced, as it was expected, which was also confirmed by statistics. Therefore, the results show that the foreseen I phrasing is fairly robust in both styles, once only 13% of the predicted I boundaries have not been produced as so, regarding intonation. Besides, just 1,4% of the predicted phonological phrase (f) boundaries (and produced as Is) were perceived as breaks by the referees. With this study one may conclude that LE and FE share the same prosodic grammar, performed by the same type of phonological/syntactic cues; nevertheless, these are more consistent in LE and have a more disperse way in FE, adding to a greater difficulty at the systematic perception of prosodic boundaries in FE than in LE.

Is “limpa” a verb or an adjective?

We investigate the role of phonological phrase boundary cues on syntactic parsing by Brazilian Portuguese (henceforth, BP) native adults. It is assumed that speech is organized in a hierarchy of prosodic constituents that may relate to constituents of other components of grammar (Nespor e Vogel, 1986). Although this is not necessarily a one-to-one relationship, a mapping is possible between constituents of each component, such as between phonological phrases and certain syntactic unit. The production of reliable prosodic cues in spontaneous speech is controversial. For instance, Snedeker e Trueswell (2003) propose that only expert speakers produce disambiguating prosodic cues; for Kraljic e Brennan (2005), even naïve speakers produce prosodic cues that are helpful for listeners. Millotte et al. (2007) found that French native speakers produced reliable prosodic cues (phrase-final lengthening and pitch rise) when they read pairs of ambiguous sentences that differed in their prosodic structures. The authors also found that native listeners were able to use these cues to assign the ambiguous words to their correct lexical categories. Then, -boundary cues may help native listeners to correctly analyze ambiguous sentences. Motivated by the French experiment results, we proposed two experiments in order to test the influence of prosody on syntactic analysis by BP adults. In the first experiment, a sentence-reading task, participants produced different prosodic patterns for ambiguous words (verb or adjective) in different syntactic structures. Duration, pitch and energy values of the segments around the-boundaries were measured and revealed that (i)-boundaries were marked by acoustical reliable cues; and (ii) the lexical categories N, V and ADJ have different behaviors in the prosodic structure. Figure A: Example of the Noun + ambiguous word - Adj [garota MUDA] (on the left) and V [garota] [MUDA…] (on the right).In the second experiment, listeners were asked to complete the auditory ambiguous sentences that were just cut after the target words (Eu acho que a menina LIMPA… . I think the clean girl…/the girl cleans…). Participants gave more verb responses in the Verb condition and more adjective responses in the Adjective condition. These results suggest that BP adults are able to use phonological phrase boundary cues to decide if an ambiguous word is a verb or an adjective and, then, to constrain syntactic analysis. We discuss the implication of these results for models of online syntactic analysis and language acquisition.Figure B: Experiment 2- Mean number of adjective and verb responses given to adjective and verb sentences (out of 4 possible responses for each sentence type).

The Prosody of Shingazidja Relatives: an update

Much work has been done in recent years on the prosody of relative clauses in Bantu languages (see among others Downing et al. 2010), and this is also the case for Shingazidja, a Bantu language of the Comoros (Patin 2010), for which it has been established that restrictive relatives differ from non-restrictive ones in that the latter, contrary to restrictives, have the relative separated from its head by a prosodic boundary, as in other languages (Cheng & Kula 2006, Cheng & Downing 2007). However, many aspects of the prosody of Shingazidja relatives remain to be established. In particular, the question of whether relatives are in this language aligned with the boundaries of Intonational phrases remained undetermined, as the H% boundary tone that characterizes these prosodic structures when they do not emerge at the end of an utterance (see O'Connor & Patin 2015) is not always observable in the data (Patin 2017). The descriptive exam of a corpus collected in 2009 indicates that a H% boundary tone does emerge at the right boundary of the relative, but that i. this tone is associated with the last surface tone and not with the last vowel, on the one hand, and ii. that it is absent from a restrictive if the restrictive relative is of reduced size, revealing that eurhythmic constraints condition the prosodic structure of these clauses.

Towards a prosodic model for Tiberian Hebrew: An intonation-based analysis

This study advances a preliminary framework for conceptualising the prosodic nature and structure of Tiberian Hebrew (TH) represented by the ṭaʿămê hammiqrāʾ through an analysis of an extant Ashkenazi cantillation tradition. The ṭaʿămê hammiqrāʾ (lit. “the senses of the reading [viz. Scripture]”) are notations added by medieval scribes to the written text of the Hebrew Bible to preserve and transmit its oral performance. Modern prosodic theory and the musical concept of conjunct and disjunct melodic motion are used to demonstrate that the ṭaʿămê hammiqrāʾ have a highly structured iconic and intonational basis that organises the system and conforms substantially to Selkirk’s (2000, 2009) cross-linguistic prosodic hierarchy. The intonation-based prosodic model proposed in this study offers a solution to the limitation Dresher (1994, 2013; see also Dresher and DeCaen 2018; DeCaen and Dresher 2020) encounters with the intonational phrase domain of his prosodic model, permitting an alternative analysis of so-called pausal forms as lengthened forms, which can occur at prosodic phrase boundaries regardless of pause. The intonation-based model is tested by assessing how accurately it reflects the cross-linguistic prosodic features of restrictive and nonrestrictive relative clauses. The results indicate that TH distinguishes three prosodic classes of relatives—prosodically marked restrictives, prosodically marked nonrestrictives, and prosodically undifferentiated relatives—findings that accord with Birkner’s (2012) intonation-based study of the prosodic structure of German relative clauses.

The Syntax of Multiple Determination in Arabic: An Anti-Residual Relative Clause/Close-Apposition Account

This paper aims to give an account of the multiple determination (determiner spreading) phenomenon in Arabic. Determiner spreading is the syntactic representation and phonological realization of multiple determiners within the same determiner phrase. As a cross-linguistic phenomenon, determiner spreading has been investigated in other languages (e.g., Scandinavian and Greek); different accounts have been proposed. For Scandinavian languages, determiner spreading has been analyzed as a representation of different semantic interpretations. As far as Greek is concerned, some analyses have been proposed; however, two prominent ones have received considerable attention in the literature: (i) a residue of a reduced relative clause and (ii) an instantiation of close appositions. Contrary to those analyses, this paper claims that none of the two analyses is suitable for Arabic; thus, a language-specific analysis is required. To analyze determiner spreading in Arabic, the current paper posits the following research question: What is the linguistic purpose of the multiple determiners found in Arabic determiner phrases? Answering the research question, the paper claims that, in addition to its indispensable role in establishing agreement between nouns and adjectives within the Arabic determiner phrase, determiner spreading demarcates syntactic and semantic phrase boundaries. The paper takes Minimalist Program and Distributed Morphology as a theoretical framework to argue that attributive adjectives are projection of an agreement phrase headed by the definite article ʔal or by the indefinite phonological marker `nunation: -n’. This proposal requires no syntactic movements in the syntax proper. The ultimate linear order is achieved in the phonological components.

Lexicon Knowledge Boosted Interaction Graph Network for Adverse Drug Reaction Recognition From Social Media.

The World Health Organization underlines the significance of adverse drug reaction (ADR) reports for patients' safety. Actually, many potential ADRs tend to be under-reported in post-market ADR surveillance. Recognizing ADRs from social media is indispensably important and could complement post-market ADR surveillance for more effective pharmacovigilance studies. However, previous approaches pose two challenges: 1) ADRs show high expression variability in social media, and thus, many potential ADRs are out-of-lexicon ones, which are difficult to be recognized, and 2) most phrasal ADRs are non-standard mentions and their boundaries are difficult to identify accurately. To tackle these challenges, we design three interaction graphs and propose a neural network approach, i.e., Interaction Graph Network (IGN). Specifically, to recognize more out-of-lexicon ADRs, besides the mentions in ADR lexicon, noun phrases in the input sentence are regarded as candidate phrases and their features are taken into considerations. Moreover, in an attempt to accurately identify ADR boundaries, three word-phrase interaction graphs are designed to represent lexicon knowledge and are encoded using graph attention networks (GATs) to directly integrate various boundary and contextual information of candidate phrases into ADR recognition. Experimental results on two benchmark datasets show that IGN can recognize ADR accurately and consistently outperforms other state-of-the-art approaches.

Acoustic and temporal representations in convolutional neural network models of prosodic events

Prosodic events such as pitch accents and phrase boundaries have various acoustic and temporal correlates that are used as features in machine learning models to automatically detect these events from speech. These features are often linguistically motivated, high-level features that are hand-crafted by experts to best represent the prosodic events to be detected or classified. An alternative approach is to use a neural network that is trained and optimized to learn suitable feature representations on its own. An open question, however, is what exactly the learned feature representation consists of, since the high-level output of a neural network is not readily interpreted. In this paper, we use a convolutional neural network (CNN) that learns such features from frame-based acoustic input descriptors. We are concerned with the question of what the CNN has learned after being trained on different datasets to perform pitch accent and phrase boundary detection. Specifically, we suggest a methodology for analyzing what temporal, acoustic and context information is latent in the learned feature representation. We use the output representations learned by the CNN to predict various manually computed (aggregated) features using linear regression. The results show that the CNN learns word duration implicitly, and indicate that certain acoustic features may help to locate relevant voiced regions in speech that are useful for detecting pitch accents and phrase boundaries. Finally, our analysis of the latent contextual information learned by the CNN involves a comparison with a sequential model (LSTM) to investigate similarities and differences in what both network types have learned.

A study of Chinese EFL learners’ problems with stress acquisition

This paper reports on a study of Chinese EFL learners' problems with the acquisition of English stress. The data were collected from 25 participants taking part in three recording tests designed by the authors. The three tests included: (1) a word list without any stress placement indicators or phonetic symbols, (2) a word list with stress placement indicators, and (3) three sentences with intonation phrase boundary marks. The results show that assigning wrong lexical stress, making the unstressed syllable almost as prominent as the stressed syllable in words, and giving inappropriate prominence to syllables in sentences are the significant factors contributing to a Chinese-English accent. Compared with previous studies on the L2 acquisition of English stress, this research offers a more detailed description of Chinese speakers' problems which have not been fully discussed so far. In addition, the authors analyze major reasons for the students' low performance, and suggest some pedagogical implications to deal with the errors detected. These implications, together with some applicable practice, will help develop the learners’ oral competence and prevent interference errors from being fossilized.

Testing the applicability of third tone sandhi at the intonation boundary

Abstract The purpose of this study is to test the applicability of Tone Three Sandhi (T3S) when the critical syllable is a monosyllabic topic preceding a topic boundary. A recitation task from 37 native speakers of Taiwan Mandarin was employed. The results from human judgements indicated that the participants predominantly produced the critical syllables with Tone 3 (T3). Additionally, the fundamental frequency of the critical syllables demonstrated a falling contour, showing that T3S was not applied. Intonation break-ups and the prolongation of the critical syllables lent strong support to the view that the topic syllable was at an intonation/phonological phrase-final position. The findings can be elegantly accommodated by constraint-based analyses, which propose that T3S must be avoided when two T3 syllables are separated by an intonation/phonological phrase boundary. Issues relating to pauses, speech rates and word frequency effects are also discussed.