Search

T3 sandhi in Mandarin is a phonological alternation where two adjacent T3-T3 tones are not allowed and become T[2]-T3. Although this phenomenon has been extensively studied in previous work, concerns about how T3 sandhi is represented in the mental lexicon remain controversial. We approached the issue of representation in a series of cross-modal priming experiments, including two semantic priming experiments with and without context. Our results indicated that T2 failed to prime a target beginning with T3 in a non-sandhi context. However, both T2 and T3 were able to induce semantic activation when appropriate contextual information was present. Conversely, when such information was absent, only T3 activated the lexical entry. Thus, our study suggests that T3 is specified, as expected for dissimilation, and T2 is resolved by a re-writing rule only under appropriate contextual information.

The author of this paper explains the interpretation of the diphthongs in the epic ”Stephen the Little” by Serbian poet Petar II Petrovich Njegos from the epoch of romanticism. They are: а) tautovocalic diphthongs in the same word, b) tautovocalic diphthongs at the word boundary and its enclitics, c) tautovocalic diphthongs at the word boundary and its proclitics, d) tautovocalic diphthongs at the boundary of two phonetic words, i.e. in sandhi. The analysis comprises the epic ”The Mountain Wreath” and the epic ”The Ray of the Microcosm”, and some of Njegos`s poems, аs well as examples from Serbian folk poems published in the first collection of folk poems ”Маlа prostonarodnya slavenо‑serbska pěsnarica” by Vuk Stefanovich Karadjich.

Phonological alternations pose challenges for models of spoken word recognition in how surface information is mapped onto stored representations in the lexicon. In the current study, an auditory-auditory priming lexical decision experiment was conducted to investigate the alternating representations of Mandarin Tone 3 in both half-third and third tone sandhi contexts. In Mandarin, a full Tone 3 (213) is reduced to an abridged tone (21) when followed by Tone 1, Tone 2, or Tone 4 (half-third tone sandhi), and Tone 3 is replaced by Tone 2 when followed by another Tone 3 (third tone sandhi). In the half-third sandhi block, disyllabic targets with a half-third (21) or full-third (213) tone FIRST syllable and a Tone 2 (35) or Tone 4 (51) second syllable were preceded by either a half-third prime, a full-third prime, or a control prime. In the third tone sandhi block, third-tone sandhi disyllabic targets with a half-third or full-third SECOND syllable were preceded by either a half-third prime, a full-third prime, or a control prime. Results showed that both half-third and full-third primes elicited significantly faster reaction times relative to the control Tone 1 condition. The size of the facilitation was not influenced by prime condition, target frequency, targets’ first syllable tone or targets’ second syllable tone. These data suggest that Mandarin T3 may be a more abstract tone and stored as the first syllable for both types of sandhi words.

The current study examined whether or not lexical access is influenced by detailed phonological features during the silent reading of Chinese sentences. We used two types of two-character target words (Mandarin sandhi-tone and base-tone). The first characters of the words in the sandhi-tone condition had a tonal alternation, but no tonal alternation was involved in the base-tone condition. Recordings of eye movements revealed that native Mandarin Chinese readers viewed the base-tone target words more briefly than the sandhi-tone target words when they were infrequent. Such articulation-specific effects on visual word processing, however, diminished for frequent words. We suggest that a conflict in tonal representation at a character/morpheme level and at a word level induces prolongation in fixation duration on infrequent sandhi-tone words, and conclude that these tonal effects appear to reflect articulation simulation of words during the silent reading of Chinese sentences.

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.

This review is based on the roundtable conference promoted by professors Carmen Matzenauer, Luciani Tenani and Christina Abreu Gomes and coordinated by professor Jose Magalhaes, aiming to discuss various perspectives in the field of phonology, in honor of the outstanding linguist Leda Bisol. A lecture given by Carmen Matzenauer on Generative Phonology and its main assumptions (e.g. levels of representation, rules and features) initiates the roundtable conference, in which she presents language acquisition data and discusses some ramifications of Generative Phonology, specifically Nonlinear Phonology. Furthermore, Luciani Tenani presents interesting examples concerning vocalic sandhi in Brazilian Portuguese based on assumptions derived from Prosodic Phonology located in the domains of Generative Phonology. Eventually, Christina Abreu Gomes discusses Exemplar-based models applied to phonological theory, referencing the continuity between lexicon and grammar proposed by these models as well as presenting effects of type and token frequencies in the use of language.

This article explores in depth various sandhi (joining) rules in Kerala’s Malayalam language, which play a vital role in framing of the inflected and agglutinated forms of words and their compounds. It discusses significant progress in a scientific method to generate a specific annotated data set of Malayalam words that would be useful in many Natural Language Processing tasks which involve Malayalam preprocessing. The article discusses the results and issues encountered in developing this word-splitting tool for Malayalam, mainly in the context of improving the alignments between parallel texts that form a core resource in the Machine Translation task.

This study is the first comprehensive acoustic study to examine the acquisition of two Mandarin tone sandhi rules: the third tone sandhi and the more phonetically motivated, half-third sandhi rule by both tonal (Cantonese) and non-tonal (American English) speakers using a Wug Test. Participants were asked to form disyllables from two monosyllabic morphemes. To test for the operation of the lexical versus the computation mechanisms in sandhi rule application, both real and various types of wug (nonsense) morphemes were included. Functional data analysis revealed that Cantonese and American speakers apply the two rules similarly on both real words and wug words, suggesting that the sandhi forms are stored as part of the representation of the abstract Tone 3 (T3) category, and computation of allophonic variants is likely to be involved during production. However, in their computation of tone sandhi rules, L2 learners showed less detailed and less accurate production of tonal contours compared to native speakers, due, perhaps, to less detailed phonological representations of allophonic variants. In general, Cantonese speakers performed better than American speakers. Perceptual mapping between Mandarin sandhi T3 to existing Cantonese tone categories may be responsible for the observed pitch contours among Cantonese speakers. Finally, no phonetic bias was found in the application of the two sandhi rules among these groups of L2 learners, which is likely due to more variability in L2’s speech, obscuring any differences that may exist.

Mandarin tone 3 sandhi is a phonological alternation in which the initial tone 3 (i.e., low tone) syllable changes to a tone 2 (i.e., rising tone) when followed by another tone 3. The present study used across-modal syllable-morpheme matchingexperiment to examine how native speakers process the sandhi sequences derived from verb reduplication and compounding, respectively. Embedded in a visually-presented sentential context, a disyllabic sequence containing a sandhi target was displayed simultaneously with a monosyllabic audio, either a tone 1 (i.e., high-level tone), tone 2 (i.e., rising tone) or tone 3 (i.e., low tone), and participants judged whether the audio syllable matched the visual morpheme. Results showed that the tone 3 sandhi was processed differently in the two constructions. The underlying tone and the surface tone were co-activated and competed with each other in sandhi compounds whereas predominant activation of the underlying tone, over the surface tone, was observed in reduplication. The processing of tone 3 sandhi offers support for distinctive morphological structures: a lexical compound is represented both as a whole-word unit and as a combination of two individual morphemes whereas a verb reduplication is represented and accessed as a monomorphemic unit in the mental lexicon.

This paper provides a corpus-based study of agreement with post-nominal adjectives in coordinate nominal expressions with only one determiner in European Portuguese (EP). It aims to provide clues to understand the patterns of agreement that contradict the predictions of our previous studies in which we made a distinction between two constructions corresponding to the same categorial linear outputs: (a) those associated with the reference of only one entity and (b) those in which nouns are used to refer distinct entities. Since it was suggested by those previous studies the existence of parallel patterns of agreement between Spanish and EP in these types of constructions, we aim to provide a preliminary comparative study that allow us to explore this possibility, namely in what concerns the closest conjunct agreement in both gender and number (considering the different behaviors of the values of these features). Finally, we aim to evaluate the adequacy of some proposals presented for Spanish to explain the EP data, contributing to the description of these structures and presenting some clues for further investigation.

Este trabalho investiga os contextos em que ocorrem processos de sândi vocálico externo em Lung’Ie, uma língua crioula de base portuguesa falada na Príncipe Island, São Tomé e Príncipe, e sua relação com o sistema suprassegmental da língua. Através da análise da relação entre acento e tom nos processos de sândi, esta pesquisa pretende demonstrar que o padrão tonal do Lung’Ie, tal como descrito por Maurer (2009), não bloqueia o fenômeno, que é bloqueado em determinados contextos de acento. A contribuição desta pesquisa envolve a compreensão do sistema suprassegmental do Lung’Ie e da relação das categorias de tom e acento na língua.

Objective: To obtain an application for iPhone mobile to identify and correct the spelling mistakes in the given Tamil text and post it in the social media. Methods: The application splits each Tamil word from the given Tamil text and further analysed by using Tamil dictionary. Each word is split morphologically and finally checked for its sandhi errors based on Tamil grammar rules. Findings: Currently, Tamil spell check application is not available for iPhone in app store. So, this will be first application for identifying misspellings in Tamil text. Applications/Improvements: This Tamil spell checker application can be improved by adding grammar rules for other languages too. Keywords: iPhone, Tamil Spell Checker, NLP

In Luganda (Bantu, Uganda), a sequence of vowels in successive syllables (V.V) is not allowed. If the first vowel is high, the two vowels are joined together in a diphthong (e.g., i + a → i͜a). If the first vowel is non-high, it is deleted with compensatory lengthening of the second vowel in the sequence (e.g., e + a → aː). This paper presents an acoustic investigation of inter-word V#V sequences in Luganda. It was found that the vowel interval in V#V sequences is longer than that in V#C sequences. When the first vowel in V#V is non-high, the formant frequency of the outcome is determined by the second vowel in the sequence. When the first vowel is high, on the other hand, the sequence is realized as a diphthong, with the transition between the two formant patterns taking up most of the duration. The durational patterns within these diphthongs provide evidence against the transcription-based claim that these sequences are reorganized so that the length lies in the second vowel (/i#V/ → [jVː]). The findings bring into question a canonical case of compensatory lengthening conditioned by glide formation.

Morphemes involved in the formation of Somali verbs and nouns are, in most instances, clearly individuated into categories corresponding to their role in word formation. Verbs contain a base, derivational extensions, inflectional affixes, and clitics that attach in a fixed order. Nouns also contain a base and derivational affixes, but little inflectional morphology. Indeed, both parts of speech have similar morphological templates in Somali, but the relationship between the language’s morphological domains and prosodic domains has only recently become a subject of detailed inquiry. We add to this ongoing trend by illustrating in this paper that there are close correlations between these domains in the language’s verbal and nominal systems that can be elucidated by morphophonological processes; certain processes occur only in a particular prosodic domain, and these process/domain combinations are similar in both the nominal and verbal systems. By establishing diagnostic phenomena attributable to phrase-level domains, this paper fills a gap between recent works focused only on defining prosodic characteristics of Somali words (Downing & Nilsson 2017; Green & Morrison 2016) and the accentual behavior of Somali clauses (Le Gac 2002, 2003a, b).

In Mandarin, a left-/right-branching asymmetry is observed when the Tone 3 Sandhi (T3S) process interacts with the syntactic structure of an expression: while expressions that have a left-branching syntactic structure only have a non-alternating sandhi pattern in which all but the rightmost T3 is changed to the sandhi tone, for expressions that have a right-branching syntactic structure various sandhi patterns are possible. This paper proposes that T3S applies cyclically bottom-up on a prosodic structure matched from the syntactic structure of an expression, along the lines of the Match Theory of syntactic-prosodic constituency correspondence (Selkirk 2011). The interaction of Match Phrase constraints and Strong Strong Start, which is a more restricted version of Selkirk’s (2011) Strong Start constraint, predicts that different prosodic structures are possible outputs for a right-branching expression, while for a left-branching expression the only possible output is a left-branching prosodic structure. The various possible sandhi patterns for a right-branching expression and the non-alternating sandhi pattern for a left-branching expression are derived when T3S applies cyclically bottom-up on the proposed prosodic structures.

This study aims to increase our knowledge about tone realiza- tion in disyllabic words in continuous Mandarin speech. Au- tomatic alignments of large speech corpora were carried out to enable the study of potential tone variants, with a special focus on variation factors such as prosodic position and right tonal context. The alignments without tone variants (V0, phono- logical representation) show that Tone 4 is more frequent in phrase-final position than in other prosodic positions, support- ing the "declination line" pattern often observed in speech pro- duction. Tone 4 is also the most frequent lexical tone (>50%) in all prosodic positions. Alignments permitting tone variants (V1, phonetic realization) show an increase of Tone 1 in phrase- initial position, compared to V0. Tone realization is observed to be related not only to the prosodic position, but also to the within-word right tonal context. Unsurprisingly, the most no- table change in tone realization happens for Tone 3 in the first syllable of disyllabic words when followed by another Tone 3 because of the well-known "tone sandhi rule" in which T3T3 disyllabic words become T2T3. Cross-word right tonal context is found to impact only Tone 3. However, the results in this study show that Tone 3 sandhi rule is more a tendency than an absolute rule.; International audience; This study aims to increase our knowledge about tone realization in disyllabic words in continuous Mandarin speech. Automatic alignments of large speech corpora were carried out to enable the study of potential tone variants, with a special focus on variation factors such as prosodic position and right tonal context. The alignments without tone variants (V0, phonological representation) show that Tone 4 is more frequent in phrase-final position than in other prosodic positions, supporting the "declination line" pattern often observed in speech production. Tone 4 is also the most frequent lexical tone (>50%) in all prosodic positions. Alignments permitting tone variants (V1, phonetic realization) show an increase of Tone 1 in phraseinitial position, compared to V0. Tone realization is observed to be related not only to the prosodic position, but also to the within-word right tonal context. Unsurprisingly, the most notable change in tone realization happens for Tone 3 in the first syllable of disyllabic words when followed by another Tone 3 because of the well-known "tone sandhi rule" in which T3T3 disyllabic words become T2T3. Crossword right tonal context is found to impact only Tone 3. However, the results in this study show that Tone 3 sandhi rule is more a tendency than an absolute rule.

Many studies have found that native Mandarin speakers produce T3 sandhi in a non-neutralized way, with the sandhi tone not fully merged into its sandhi target. This makes it interesting to investigate whether advanced learners of Mandarin will produce T3 sandhi in a similar or different way as well as reasons underlying it. Results from this carefully controlled experimental study show that advanced Korean learners of Mandarin produce T3 sandhi in a neutralized way, in contrast to non-neutralized production by native Mandarin speakers. This is probably due to Korean speakers’ learning of the explicit T3 sandhi rule in their L2 learning program, as compared to the implicit learning of the T3 sandhi rule by Mandarin speakers. Though Korean speakers produce T3 sandhi in a neutralized way, their production of sandhi T3 is found to be overall flatter in f0 curve than that of native Mandarin speakers; this is consistent with their production of Mandarin lexical T2, the base tone of the sandhi target T2. This finding suggests that these advanced L2 learners’ sandhi production is restrained by the way they produce Mandarin lexical tones.

This paper describes neural network based approaches to the process of the formation and splitting of word-compounding, respectively known as the Sandhi and Vichchhed, in Sanskrit language. Sandhi is an important idea essential to morphological analysis of Sanskrit texts. Sandhi leads to word transformations at word boundaries. The rules of Sandhi formation are well defined but complex, sometimes optional and in some cases, require knowledge about the nature of the words being compounded. Sandhi split or Vichchhed is an even more difficult task given its non uniqueness and context dependence. In this work, we propose the route of formulating the problem as a sequence to sequence prediction task, using modern deep learning techniques. Being the first fully data driven technique, we demonstrate that our model has an accuracy better than the existing methods on multiple standard datasets, despite not using any additional lexical or morphological resources. The code is being made available at https://github.com/IITD-DataScience/Sandhi_Prakarana

Tone 3 (T3) in Mandarin is one of the most difficult tones for second-language (L2) learners given its variants in different contexts. While previous studies investigated L2 learners’ acquisition of T3 in the sandhi context, it remains unclear how L2 learners’ learning experience modulates their phonological knowledge of T3 sandhi in producing different types of words. This study used a wug production test to investigate the effect of Mandarin learning experience on the production of the sandhi form (a rising tone) by experienced and inexperienced Korean-speaking learners. The acoustic analyses showed that experienced Korean-speaking L2 learners were better at using their phonological knowledge of T3 sandhi than less experienced learners when producing pseudo and novel words, but not real Chinese words. The experienced learners had higher pitch values and a steeper rising slope for the T3 sandhi form of pseudo words, and higher pitch values for that of novel words, than the inexperienced learners. The findings suggested that learners’ increased experience with Mandarin facilitated their use of the phonological knowledge of T3 sandhi. Given learners’ difficulties with the T3 sandhi rule, language teachers are suggested to develop teaching materials with different word types to promote the generalization of the rule.

This study investigated the perception of Mandarin tonal alternations in disyllabic words. In Mandarin, a low-dipping Tone3 is converted to a high-rising Tone2 when followed by another Tone3, known as third tone sandhi. Although previous studies showed statistically differences in F0 between a high-rising Sandhi-Tone3 (T3) and a Tone2, native Mandarin listeners failed to correctly categorize these two tones in perception tasks (Peng, 2000). The current study utilized the visual-world paradigm in eye-tracking to further examine whether acoustic details in lexical tone aid lexical access in Mandarin. Results showed that Mandarin listeners tend to first process Sandhi-T3 as Sandhi-T3 and Tone2 as both tones, then later detect the acoustic differences between the two tones revealed by the sandhi context, and finally activate the target word during lexical access. The eye-tracking results suggest that subtle acoustic details of F0 may facilitate lexical access in automatic fashion in a tone language.

Language-specific and context-dependent phonological rules of lexical tone are prevalent in tone languages. Such rules are commonly referred to as tone sandhi. One of the most studied sandhi rules is Mandarin Tone 3 sandhi. In Mandarin, Tone 3 followed by another Tone 3 is pronounced as Tone 2 (33 → 23). In this chapter, we reviewed our current understanding of the processing of Tone 3 sandhi. Two important and relatively well-investigated questions are whether Tone 3 sandhi involves on-line tone substitution in speech production and whether the auditory representations of Tone 2 and Tone 3 are less distinct from each other due to the acquisition of Tone 3 sandhi. Recent behavioral studies demonstrated that in the lexical decision task, only Tone 3 had a facilitation effect on targets carrying tone sequence 33, while in the picture-naming task, a facilitation effect was found with both Tone 2 and Tone 3. These results supported that Tone 3 sandhi involves on-line tone substitution, in line with fMRI studies showing that Tone 3 sandhi resulted in higher activation in the right pIFG, which is known to engage in articulatory representations and their sequencing. Regarding tone perception, previous behavioral studies showed that the acquisition of Tone 3 sandhi led to worse performance at discriminating Tone 2 and Tone 3. Further, the contrast between Tone 2 and Tone 3 is consistently reported to elicit reduced MMN compared to other tone pairs only in native speakers. One explanation of these findings is that the auditory representations of Tone 2 and Tone 3 activated each other due to Tone 3 sandhi. Namely, high-level phonological rule could modulate pre-attentive auditory processing. In the future, the role of linguistic context in the processing of tone sandhi needs more investigation, especially regarding how listeners retrieve the correct word/morpheme based on the contextual information.

Through this paper we are proposing a framework for enunciation of Sanskrit shlokas and phrases. Sanskrit shlokas have been known to have a cathartic effect on the mind and soul. But due to the richness of the language and the lack of a large number of teachers, few are able to tap its benefits. We propose a new method of sandhi splitting based on the shloka characteristics and the utilization of the seq2seq model with attention for generating shlokas. On testing, the model generates an MOS of 3.25 out of 4.

Este artigo tem como objetivo analisar a ocorrência de processos fonológicos de cunho estilístico a partir de um corpus poético do Português Arcaico, as Cantigas de Santa Maria, atribuídas à lavra do Rei Afonso X de Castela, o rei “Sábio”. Trata-se, portanto, de verificar os possíveis condicionamentos linguísticos envolvidos na sua realização, sobretudo no que se refere a fenômenos de sândi – crase, elisão e hiato, conforme a noção de opacidade proposta por Kiparsky (1985). Com base nos dados analisados, foi possível concluir que o trovador não possui tanta liberdade para “criar” ou não formas desviantes, segundo sua vontade, para satisfazer necessidades artísticas, fato corroborado pela baixíssima margem de arbítrio de que dispunham.

This study probes the phonetic and phonological knowledge of lexical tones in TTS models through two experiments. Controlled stimuli for testing tonal coarticulation and tone sandhi in Mandarin were fed into Tacotron 2 and WaveGlow to generate speech samples, which were subject to acoustic analysis and human evaluation. Results show that both baseline Tacotron 2 and Tacotron 2 with BERT embeddings capture the surface tonal coarticulation patterns well but fail to consistently apply the Tone-3 sandhi rule to novel sentences. Incorporating pre-trained BERT embeddings into Tacotron 2 improves the naturalness and prosody performance, and yields better generalization of Tone-3 sandhi rules to novel complex sentences, although the overall accuracy for Tone-3 sandhi was still low. Given that TTS models do capture some linguistic phenomena, it is argued that they can be used to generate and validate certain linguistic hypotheses. On the other hand, it is also suggested that linguistically informed stimuli should be included in the training and the evaluation of TTS models., Submitted to Speech Prosody 2020

Morphological analysis is one of the fundamental tasks in computational processing of natural languages. It is the study of the rules of word construction by analysing the syntactic properties and morphological information. In order to perform this task, morphemes have to be separated from the original word. This process is termed as sandhi splitting. Sandhi splitting is important in the morphological analysis of agglutinative languages like Malayalam, because of the richness in morphology, inflections and sandhi. Due to sandhi, many morphological changes occur at the conjoining position of morphemes. Therefore, determining the morpheme boundaries becomes a tough task, especially in languages like Malayalam. In this paper, we propose a deep learning approach for learning the rules for identifying the morphemes automatically and segmenting them from the original word. Then, individual morphemes can be further analysed to identify the grammatical structure of the word. Three different systems were developed for this analysis using Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU) and obtained accuracies 98.08%, 97.88% and 98.16% respectively.

This study illustrates how three level tones might have developed diachronically by comparing two synchronic Mandarin dialects. In Standard Mandarin (SM), the four lexical tones are denoted as /H, LH, L, HL/ or /H, R, L, F/ phonologically. However, based on evidence from two acoustic experiments, this study proposes that the four lexical tones in Taiwan Mandarin (TM) should be analyzed as /H, M, L, HM/, with /H, HM/ in a high register and /M, L/ in a low register. The proposed tonal structure can account for all the tone sandhi in TM using the framework of Optimality theory, and the register difference plays an important role in the analyses. Also, the new TM tonal representation has an advantage in explaining the absence of the SM Tone 2 Sandhi. The new tonal representations illuminate how pitch contour differences might have developed into structural tone changes.

This paper intends to investigate Greek influence on the Latin sound change [b] > [β] suggested occasionally in the literature by surveying not only the relevant linguistic data of Latin/Romance and Koine/Modern Greek but also the relevant literature and by involving and analyzing data sets recorded from 18 Roman provinces and the city of Rome in the Computerized Historical Linguistic Database of the Latin Inscriptions of the Imperial Age (cf. http://lldb.elte.hu/) by a more differentiated phonological approach considering external sandhi rules and in a chronological distribution more detailed than any applied before. In the end, the influence of Greek has been evidenced at least for some areas and especially for the early period (1 st –3 rd century AD), which is more important in this respect than the late period (4 th –6 th century AD), since then the merger can also be explained by developments in Latin itself beside a supposed external influence.

Parts of speech (POS) tagging is the process of assigning the part of speech tag to each and every word in a sentence. In this paper, we have presented POS tagger for Kannada, a low resource south Asian language, using Condition Random Fields. POS tagger developed in the work uses novel features native to Kannada language. The novel features include Sandhi splitting, where a compound word is broken down into two or more meaningful constituent words. The proposed model is trained and tested on the tagged dataset which contains 21 thousand sentences and achieves a highest accuracy of 94.56%.

This article traces the history of how modern Western linguistics adopted the term 'sandhi' from the Sanskrit grammatical tradition and adapted it to its theoretical needs. In particular, we will acknowledge the fundamental role played by Müller,1 who combining both Indic ('Prakriyā' grammars and 'Prātiśākhyas') and Western approaches (those of Colebrooke and Bopp) to the representation of Sanskrit grammar, coined in 1866 the labels of 'internal sandhi' and 'external sandhi'. Such labels gained momentum thanks to the works of Whitney in the 19th century and Bloomfield in the 20th century and eventually became common parlance in Western linguistics.

This paper highlights a hitherto unreported change in progress among northern speakers of British English, with increasing post-nasal [ɡ]-presence in words like 'sing' or 'wrong' pre-pausally. The factors that condition this innovation are unclear due to collinearity between various boundary phenomena. The right edge of phrasal prosodic categories may be associated with boundary tones, final lengthening, and pause; consequently, the variable presence of [ɡ] appears to be affected by prosodic boundary strength, segmental duration, and the presence and duration of a following pause. These factors are teased apart through analysis of an elicitation task from 30 northern speakers, which reveals that [ŋɡ] clusters are conditioned most strongly by pause. Post-nasal [ɡ]-presence is only licensed when the following consonant-initial word is temporally distant, showing only minimal sensitivity to prosodic boundaries directly. The surface effect of segmental duration arises only indirectly through its collinearity with pause duration. Current theoretical approaches to external sandhi emphasize a range of different factors, including phonological representations of prosodic constituency, phonetic parameters like segmental duration, and psycholinguistic mechanisms of production planning. This paper provides quantitative evidence from an under-reported feature of northern English that bears directly on these debates.

This work explored how the effects of input modalities on the production of different L2 contrasts may change over the course of L2 phonological development. Spontaneous imitation and orthography reading tasks were framed within a longitudinal investigation tracking the acquisition of Mandarin stops and lexical tones by Korean learners developing from a near-naive to a novice stage. Imitation prompted by identical auditory stimuli across different phases revealed a decline in the production of T3, which was produced as T2 to a greater degree with more Mandarin experience. This somewhat counterintuitive finding suggests that the initial benefits from the salient acoustic properties of T3 were overridden by the knowledge of the L2 sound system, namely T3 being realized as T2 via T3 sandhi, leading to diminishing T2-T3 contrasts over time. The stop production data, on the other hand, demonstrated the overarching impact of the existing grapheme-to-phoneme association on the phonological development of new L2 sounds. Some learners prompted with Pinyin produced Mandarin unaspirated stops (e.g., ) with long aspiration and low-F0 for high-level tones. This phonetically unfaithful mapping arose from the familiarity with English voiced stops that are often adapted as lenis stops [long VOT – low F0] by Korean learners. Taken together, the findings demonstrate that the effects of input modalities are specific to the different stages of early L2 phonological acquisition as well as to the types of L2 contrasts.

In Hanoi Vietnamese, the rising and falling tones are frequently confused before the (high) level tone, even though they are clearly distinct in other contexts. In this paper, we conduct production and perception experiments designed to assess the source of this confusion. We argue that the peak of the rising tone is normally delayed onto the initial portion of the following tone, but that this peak delay lacks acoustic and perceptual salience when this following tone is a level tone. As a result, the rising tone before a level tone is often perceived as a falling tone. Although we rule out the possibility that the tonal pattern under investigation is a phonological alternation, we propose that the complex coarticulatory and perceptual mechanisms that underlie it could account for the development of other instances of regressive tone sandhi.

The f0 realization of lexical tones is known to vary greatly due to contextual tonal variation processes such as tone sandhi and tonal coarticulation. This study set out to investigate the f0 variability induced by these two different sources with well-controlled acoustic data from Tianjin Mandarin. Tianjin Mandarin is known for its complex patterns of tonal variation over disyllabic constituents, as well as for the conflicting directionality of sandhi alternations and possible iterative application of disyllabic sandhi rules over trisyllabic constituents. Previous studies on Tianjin tone sandhi have often been based on impressionistic descriptions of speech produced by speakers of older generations. In this study, we investigated the f0 realization of disyllabic and trisyllabic tonal sequences in Tianjin Mandarin produced by younger speakers (born in the 1980s). Through examining f0 variation of lexical tones as a function of different following tones, we observed an interesting anticipatory raising effect of the lexical low-falling tone (T1). Furthermore, we confirmed three tonal sequences with tone sandhi changes: (1) two low-falling tones (T1T1), (2) a high-falling tone followed by a low-falling tone (T4T1), and (3) two low-dipping tones (T3T3). These disyllabic tone sandhi patterns, however, were not consistently observed within trisyllabic sequences, as claimed in the literature. Specifically, while T3T3 showed sandhi application regardless of its position within a trisyllabic sequence, T1T1 and T4T1 sequences showed sandhi changes only when they were right aligned. Last but not least, no tonal neutralization was observed over any of the sandhi processes. Rather, we argue that T3T3 is a near-merger sandhi case, while T1T1 and T4T1 can be classified as no-merger sandhi sequences.

The morphological richness and the agglutinative nature of Malayalam make it necessary to retrieve the root word from its inflected form in most of the NLP tasks. This paper presents an approach to identify the suffixes of Malayalam words using MBLP approach. The idea here is to use Memory Based Language Processing (MBLP) algorithm for Malayalam suffix identification. MBLP is an approach to language processing based on exemplar storage during learning and analogical reasoning during processing. Sandhi splitting is essential for morphological analysis, document indexing and topic modeling. Suffix separation improves the quality of machine translated text. Training instances created from words are manually annotated for their segmentation and the system is trained using TiMBL (Tilberg Memory Based Learner). The paper presents memory-based model of Malayalam suffix identification and its generalization accuracy.

Studies on Mandarin tone sandhi have focused on Beijing Mandarin. Taiwan has been politically separated from mainland China since 1949, but it is not known if tone sandhi in Taiwan Mandarin displays different patterns or characteristics. However, there has been no comparative study to investigate if Beijing Mandarin and Taiwan Mandarin display the same tone sandhi pattern. This study aims to fill this gap by comparing Beijing and Taiwan Mandarin through a productive experiment to examine acoustic differences between sandhied tone 3 and lexical tone 2. The results indicate that tone 3 sandhi among Mandarin dialects is not a homogeneous category, but displays a graded phenomenon of a categorical change and tonal reduction. The experimental evidence shows that acoustic difference between sandhied tone 3 and lexical tone 2 is larger in Beijing Mandarin than that in Taiwan Mandarin. Gender effects are also detected and acoustic difference in female samples is consistently larger than that in male samples across Beijing and Taiwan Mandarin. The findings suggest that the third tone sandhi in Beijing Mandarin is more like a categorical change (i.e., changed to lexical tone 2) whereas the sandhi in Taiwan Mandarin is more like a tonal reduction.

It is well-known that in Standard Chinese, the sandhi-derived T3 is realized with a risingf 0 pattern that is typically identical with the lexical T2. However, it is less understood on how dialect backgrounds play a role in T3 sandhi perception. The present study set out to shed some light on this issue with evidence with eye movements by examining the participants’ looks to both targets and competitors within the Visual World Paradigm when they heard T3T3 stimuli as well as their T2T3 near-minimal-pair counterparts. We recruited participants from both Beijing and Nanjing. Our results show that, due to the modulation of their native tonal system, Nanjing listeners showed clear early advantages in the online processing of tones in Standard Chinese, as reflected with more looks to the target as compared to the Beijing group.

The current study aims to investigate the categoricalness versus gradience of tone sandhi application in assimilatory and dissimilatory tone sandhi processes, within the setting of one language. Our hypotheses were that (a) assimilatory processes can be gradient, based on their articulatory motivation; and (b) dissimilatory processes should always be categorical, hence never show any gradient application, based on Ohala’s “hypercorrection” theory. We selected in Nanjing Mandarin two pairs of comparable assimilatory and dissimilatory tone sandhi processes based on previous researchers’ observations. Results show a near-categorical assimilatory Sandhi 1 and a gradient assimilatory Sandhi 3, congruent with the prediction that assimilatory processes are allowed to apply in a gradient fashion. Though we found that dissimilatory Sandhi 4 is a categorical process, our observation of gradience in dissimilatory Sandhi 2 suggests that dissimilatory changes can also occur in a gradient way, which contradicts the prediction by the “hyper-correction” account.

This study examines the distribution of creaky voice as a function of various prosodic structures in a large-scale corpus of continuous speech of Mandarin. Both tonal categories and prosodic boundaries have strong effects on the likelihood of creak and relative creakiness. It was found that (1) creaky voice in Mandarin is indeed largely driven by the occurrence of low pitch and weakening; (2) Tone 3 sandhi and Tone 2 are different in both pitch and voice quality; (3) the creakiness of Tone 3 (low tone) and the neutral tone (weakening) is realized differently. Moreover, females and males do not differ in how they use creaky voice linguistically.

Malayalam is a highly agglutinated resource-poor language. A lot of words in Malayalam are formed by the repeated addition of suffixes into root forms of the words called stems. Such words are called compound words. Identifying root words from the compound words is a key challenge in the preprocessing of Malayalam text. The grammar rules in Malayalam known as sandhi rules play a vital role in the splitting and formation of compound words. In this paper, we have proposed a hybrid methodology for suffix separation in Malayalam compound words. The sandhi rule associated with each compound word is recognized through machine learning approach. And suffix separation is performed using a rule-based approach. Since a lot of rules are required for suffix separation in Malayalam, we have only considered ‘Agamasandhi’ words for rule-based splitting. The proposed system provides better results and can be extended to prepare a complete suffix separation system for Malayalam.

Sixth International Symposium on Tonal Aspects of Languages, 18-20 June 2018, Berlin, Germany

Large numbers of children around the world are learning tone languages, but few studies have examined the acoustic properties of children's early tone productions. Even more scarce are acquisition studies on tone sandhi, a tone change phenomenon which alters the surface realization of lexical tones. Two studies using perceptual coding report the emergence of lexical tone and tone sandhi at around 2 years (Li and Thompson, 1977; Hua and Dodd, 2000). However, the only acoustic study available shows that 3-year-olds are not yet adult-like in their lexical tone productions (Wong, 2012). This raises questions about when children's productions become acoustically adult-like and how their tone productions differ from those of adults. These questions were addressed in the current study which compared Mandarin-speaking pre-schoolers' (3–5-year-olds) tone productions to that of adults. A picture naming task was used with disyllabic real words familiar to pre-schoolers. Overall children produced appropriate tone contours for all tones, i.e., level for tone 1, rising for tones 2, 3 and full sandhi, falling for tone 4 and half sandhi. However, children's productions were not adult-like for tones 3, 4, and the sandhi forms, in terms of coordinating pitch range, slope and curvature, with little evidence of development across ages. These results suggest a protracted process in achieving adult-like acoustic realization of both lexical and sandhi tones.

The texts in Olden days were written in a concise style known as Sutra style. This was meant especially for the brevity and for the oral transformation of the knowledge. Many devices of writing like Tantrayukti, Tacchilya, Arthashraya and Nyayah etc. were used to explain the complex matter in a simple manner and also to preserve the knowledge from improper persons. Sanskrit scholars have enriched and embellished Sanskrit language by various devices like Chandas, Alankara, Sandhi, Samasa, Tantrayukties, Nyayas, Tacchilyas, Arthashrayas, Kalpana etc. among which the Nyayas have a special place which explains the complex subject in simple manner by using the common proverbs and sayings. This is one of the devices by which an experience secured from or a conclusion reached in a particular case can be used to explain a similar situation in a brief and effective manner. A Nyayah may be used to explain a similar situation in a brief and to explain the situation in a given case forcefully. It was therefore natural that Ayurvedic Literature was also influenced by these maxims whenever faced with any ambiguity or conflict in the Shastras and for the sake of brevity maxims were used by the authors. So this present article highlights the maxims used and also to explore the particular circumstances where these maxims are used in the inscription of the Ayurvedic literature

This chapter reviews the significant development of research on the second language acquisition of the sound system of standard Chinese. While major findings from studies on both segmental (vowels and consonants) and supra-segmental features of L2 Chinese phonology are outlined, particular attention is paid to the acquisition of Chinese lexical tones. This review begins with a summary of how non-native tonal errors have historically been viewed and how these views have changed over time. Several aspects of the second language Chinese tone studies are reviewed and discussed, such as the acquisition order of tones, the third tone and its sandhi rules, tone perception and production, positional effects in second language tone grammar and the acquisition of Chinese intonation, etc., followed by a discussion regarding research approaches of empirical studies of the second language perception and production of Chinese tones. The chapter ends with pedagogical implications and points out future directions of research on second language Chinese sounds.

The paper introduces end-to-end neural network models that tokenize Sanskrit by jointly splitting compounds and resolving phonetic merges (Sandhi). Tokenization of Sanskrit depends on local phonetic and distant semantic features that are incorporated using convolutional and recurrent elements. Contrary to most previous systems, our models do not require feature engineering or extern linguistic resources, but operate solely on parallel versions of raw and segmented text. The models discussed in this paper clearly improve over previous approaches to Sanskrit word segmentation. As they are language agnostic, we will demonstrate that they also outperform the state of the art for the related task of German compound splitting.

In Sanskrit, small words (morphemes) are combined to form compound words through a process known as Sandhi. Sandhi splitting is the process of splitting a given compound word into its constituent morphemes. Although rules governing word splitting exists in the language, it is highly challenging to identify the location of the splits in a compound word. Though existing Sandhi splitting systems incorporate these pre-defined splitting rules, they have a low accuracy as the same compound word might be broken down in multiple ways to provide syntactically correct splits. In this research, we propose a novel deep learning architecture called Double Decoder RNN (DD-RNN), which (i) predicts the location of the split(s) with 95% accuracy, and (ii) predicts the constituent words (learning the Sandhi splitting rules) with 79.5% accuracy, outperforming the state-of-art by 20%. Additionally, we show the generalization capability of our deep learning model, by showing competitive results in the problem of Chinese word segmentation, as well., Comment: Accepted in EMNLP 2018