Search

Over the past decade, learner corpora have gained recognition as valuable data sources in Second Language Acquisition (SLA) research. This development can be attributed to significant progress in Learner Corpus Research (LCR). However, there is still substantial work to be done. This article highlights key issues essential for sustaining the relevance of learner corpora in SLA. More particularly, I focus on the need for more diverse types of learner corpora, stress the importance of detailed metadata, and advocate for multifactorial study designs. I then revisit ongoing debates regarding the role of the native speaker in LCR and propose a practical solution to address this thorny issue. Finally, I also readdress the need for improvement in the quantitative methods and statistics, arguing that the importance of robust quantitative analysis cannot be overstated. In conclusion, I envision an ambitious learner corpus compilation project that adheres to the FAIR principles, with the goal of further elevating study quality in LCR. [ABSTRACT FROM AUTHOR]

This paper describes the rationale for and design of a new multimodal corpus of L2 academic English from a Sino-British university in China: the Corpus of Chinese Academic Written and Spoken English (cawse). The unique context for this corpus provides language samples from Chinese students who use English as a second language (L2) in a preliminary-year programme, which prepares students for academic studies at university level, at a campus where English is used as the Medium of Instruction (emi). Data were collected from a variety of settings, including written (i.e., exam scripts and essays) and spoken assessments (i.e., interviews and presentations), covering the full range of grades awarded to those language samples, as well as from student group interactions during teaching and learning activities. The multimodal nature of the corpus is realised through the availability of selected audio/video recordings accompanied by the orthographically transcribed text. This open-access corpus is designed to help shed light on Chinese students' academic L2 English language use in a variety of written, spoken and multimodal discourses. [ABSTRACT FROM AUTHOR]

We present the JVNV, a Japanese emotional speech corpus with verbal content and nonverbal vocalizations whose scripts are generated by a large-scale language model. Existing emotional speech corpora lack not only proper emotional scripts but also nonverbal vocalizations (NVs) that are essential expressions in spoken language to express emotions. We propose an automatic script generation method to produce emotional scripts by providing seed words with sentiment polarity and phrases of nonverbal vocalizations to ChatGPT using prompt engineering. We select 514 scripts with balanced phoneme coverage from the generated candidate scripts with the assistance of emotion confidence scores and language fluency scores. Experimental results show that JVNV has better phoneme coverage and emotion recognizability than previous Japanese emotional speech corpora. We then benchmark JVNV on emotional text-to-speech synthesis using discrete codes to represent NVs. The results demonstrate that there still exists a gap between the performance of synthesizing read-aloud speech and emotional speech, and adding NVs in the speech makes the task even harder, which brings new challenges for this task and makes JVNV a valuable resource for relevant works in the future. To our best knowledge, JVNV is the first speech corpus that generates scripts automatically using large language models.

The performance of speech-processing models is heavily influenced by the speech corpus that is used for training and evaluation. In this study, we propose BAlanced Script PROducer (BASPRO) system, which can automatically construct a phonetically balanced and rich set of Chinese sentences for collecting Mandarin Chinese speech data. First, we used pretrained natural language processing systems to extract ten-character candidate sentences from a large corpus of Chinese news texts. Then, we applied a genetic algorithm-based method to select 20 phonetically balanced sentence sets, each containing 20 sentences, from the candidate sentences. Using BASPRO, we obtained a recording script called TMNews, which contains 400 ten-character sentences. TMNews covers 84% of the syllables used in the real world. Moreover, the syllable distribution has 0.96 cosine similarity to the real-world syllable distribution. We converted the script into a speech corpus using two text-to-speech systems. Using the designed speech corpus, we tested the performances of speech enhancement (SE) and automatic speech recognition (ASR), which are one of the most important regression- and classification-based speech processing tasks, respectively. The experimental results show that the SE and ASR models trained on the designed speech corpus outperform their counterparts trained on a randomly composed speech corpus. [ABSTRACT FROM AUTHOR]

ORAL and ORTOFON, general corpora of the spoken Czech language, capture authentic and prototypical informal spoken language. DIALEKT, a specialized corpus, represents traditional regional dialects of the Czech language. Since the corpora's goals and the nature of the captured language data differ, different data collection methods were required. It concerns not only the choice of speakers, but the whole communication situation. Samples chosen from these three corpora are included in the Mapka application and reflect the distinct character of the corpora. The ORAL and ORTOFON samples show general spoken language in various informal situations and capture a wide range of speakers. The DIALEKT samples represent traditional regional dialects spoken by chosen types of speakers in a semiformal situation of guided interview. [ABSTRACT FROM AUTHOR]

According to data of UNESCO, in 2013, Latgalian language with 150,000 users was recognised as one of the world’s endangered and vulnerable languages, as all generations still use the oral form, but the sustainability of the language is seriously jeopardised, since the number of young language users decreases. Pursuant to the EU directives and recommendations for preservation, research and development of regional and endangered languages, as well as the Guidelines for the State Language Policy 2021–2027 regarding development, disclosure on the web and accessibility of varied text corpus, in 2020, a group of researchers of the Rēzekne Academy of Technologies in the Project of State Research Programme Digital Resources of Humanities: Integration and Development (No. VPP-IZM-DH-2020/1-0001) started its work on the development of the Contemporary Latgalian Speech Corpus (MuLaR) aimed at the documentation, research, studies and acquisition of Latgalian. The aim of the article is to identify and analyse the issues that are important in the process of creating MuLaR, applying the referential analysis of the scientific literature and comparative methodology. In turn, applying the analytical-synthetic method and based on the experience accumulated by the corpus creators, there was developed an initial model for the corpus architectonics and technological solutions, covering such issues as ensuring a representative Latgalian speech corpus, bearing in mind the territorial distribution of Latgalian language communities and diversity of Latgalian patois; the most appropriate methods to document natural, spontaneous language: collection of new data, opportunities to use the existing recordings (interviews, TV, radio broadcasts, field research data collections), other databases (reiti.rta.lv); understanding metadata; ethical aspects of the speech corpus; transcribing (software, conventions to reveal the features of spoken text as accurately as possible); creation of an accessible, easy-to-use open-access platform, using the experience of creating oral speech corpuses for lesser-used languages / dialects in other countries. The article declares the main challenges for the corpus development after the initial validation of the corpus data, including in relation to the morphological tagging possibilities of the corpus. [ABSTRACT FROM AUTHOR]

Copyright of CIRCULO de Linguistica Aplicada a la Comunicacion is the property of Universidad Complutense de Madrid and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Despite the increasingly wide-ranging accessibility of L2 Chinese learner corpora and achievements in Chinese Learner Corpus Research (CLCR), there are no studies which provide a critical analysis of the key features and limitations of this expanding field. This study therefore aims to fill this lack in the literature by both investigating the actual state of CLCR and existing L2 Chinese learner corpora and outlining their main features and critical issues. First, the paper introduces the development and current trends in CLCR, particularly emphasizing the widespread use of multi-method approaches in this field. Second, it focuses on design issues of L2 Chinese learner corpora, identifying main characteristics and limitations. The paper shows that L2 Chinese learner corpora present many gaps concerning language-related, task-related, and learner-related criteria. Gaps in learner corpus analysis and annotation are also discussed. Third, the paper offers the first analysis conducted to date that categorizes existing L2 Chinese learner corpora according to mode (written, spoken and multimodal) and size (largescale and small-scale). Finally, the article directs attention toward challenges in this field, concluding with future directions for CLCR and its intersections with Second Language Acquisition (SLA) to support L2 Chinese teaching and learning. Suggestions on direct and indirect applications of L2 Chinese learner corpora data are also provided. [ABSTRACT FROM AUTHOR]

This paper presents a synchronic and diachronic computer corpus of Makarska littoral dialects. This corpus was created as part of the project to explore the ikavian neoštokavian dialects of the narrow coastal area in Croatian region of Dalmatia around the town of Makarska. The dialectological characteristics of the dialects studied are briefly presented first, followed by presentation of the digital system. The system is logically organized in first part as a corpus of literary texts created from 1729 to 1803 and digitally processed, and in the second part from the materials collected through dialectological questionnaires prepared and methodologically adapted as part of the creation of the Croatian Linguistic Atlas. Methods of collecting linguistic data, method of input into the digital form and methods and possibilities of data processing will be explained. Based on the input and search strategies within the system, the examples will prove the origin of the dialects of the Makarska littoral to be that of the ikavian neoštokavian dialect described in the dialectological literature. This computer-based principle of work is a novelty in Croatian dialectology which has not been digitally processed so far and offers a basis for future dialectological research. This platform can be used in order to shorten the time of data processing and to analyse them more systematically and more efficiently. So far, there has been no such digital repository for any Croatian speech. This project represents a thorough synchronic and diachronic study of one rounded language area. [ABSTRACT FROM AUTHOR]

Malaysian English (ME) has received much research attention in terms of its linguistic system, and use. In the last two decades, with the development of language corpora and corpus methods, research in ME as an Outer Circle and postcolonial variety increased significantly. These corpus-based studies are important and indeed interesting but they are limited to descriptions and discussions on contemporary ME, as they are based on synchronic ME corpora. Research in diachronic changes in ME, to date, is rare to say the least. To address this gap and facilitate systematic examination of changes in ME necessitates the development of diachronic ME corpora. This article reports on the design and development of the first diachronic Malaysian English corpus (DMEC), comprising Malaysian English written texts from the 1960s until the 2010s. The six decades represent three phases of English in Malaysia, end of colonial era, postcolonial and contemporary ME. The corpus is designed to facilitate research in ME changes in terms of its linguistic system, identity and trajectory. Besides the design and development of the DMEC, three sample analyses based on the existing corpus are discussed to demonstrate the value of the corpus as a resource for diachronic research in ME. The current article, overall, contributes to existing knowledge on the development of language corpora in Malaysia and foregrounds the potential of diachronic research in ME. [ABSTRACT FROM AUTHOR]

This paper discusses the creation and use of the Coronavirus Corpus, which is currently (March 2021) 900 million words in size, and which will probably be about one billion words in size by May–June 2021. The Coronavirus Corpus is a subset of the NOW Corpus (News on the Web), which is currently about 12.1 billion words in size and which grows by about two billion words each year. These two corpora are updated every night, with about 6–10 million words for NOW and 2–3 million words for the Coronavirus Corpus. The Coronavirus Corpus allows users to see the frequency of words and phrases over time (even by individual day), and users can find all words that are more frequent in one time period than another. Users can also see the collocates for words and phrases, and compare the collocates to see what is being said about particular topics over time. [ABSTRACT FROM AUTHOR]

This paper aims to design and validate a phonetically balanced speech corpus for Arabic language. Designing and developing a rich and phonetically balanced corpus in optimal context is one of the key issues in building high quality of text-to-speech synthesis systems. The rich characteristic is in the sense that it must contain all the possible phonemes on the right and left context, whereas the balanced characteristic is in the sense that it respects the phonetic distribution in the language. We propose a new methodology for designing and implementing such corpus for speech synthesis purposes. The paper explains the whole creation process of this corpus, beginning with the design stage, corpus creation, recording phases, and finally the segmentation of the speech corpus. The speech corpus contains 202 sentences with 6174 phonemes. In order to validate the speech corpus, an Arabic speech synthesis system using Hidden Markov Models has been developed. [ABSTRACT FROM AUTHOR]

This article presents research on the detection of pathologies affecting speech through automatic analysis. Voice processing has indeed been used for evaluating several diseases such as Parkinson, Alzheimer, or depression. If some studies present results that seem sufficient for clinical applications, this is not the case for the detection of sleepiness. Even two international challenges and the recent advent of deep learning techniques have still not managed to change this situation. This article explores the hypothesis that the observed average performances of automatic processing find their cause in the design of the corpora. To this aim, we first discuss and refine the concept of sleepiness related to the ground-truth labels. Second, we present an in-depth study of four corpora, bringing to light the methodological choices that have been made and the underlying biases they may have induced. Finally, in light of this information, we propose guidelines for the design of new corpora.

Any expansion of the TEI beyond its traditional user base involves a recognition that there are many differing answers to the traditional question “What is text, really?” We report on some work carried out in the context of the COST Action Distant Reading for European Literary History (CA16204), in particular on the TEI-conformant schemas developed for one of its principal deliverables: the European Literary Text Collection (ELTeC). The ELTeC will contain comparable corpora for each of at least a dozen European languages, each being a balanced sample of one hundred novels from the period 1840 to 1920, together with metadata concerning their production and reception. We hope that it will become a reliable basis for comparative work in data-driven textual analytics. The focus of the ELTeC encoding scheme is not to represent texts in all their original complexity, nor to duplicate the work of scholarly editors. Instead, we aim to facilitate a richer and better-informed distant reading than a transcription of lexical content alone would permit. At the same time, where the TEI encourages diversity, we enforce consistency by permitting representation of only a specific and quite small set of textual features, both structural and analytical. These constraints are expressed by a master TEI ODD, from which we derive three different schemas by ODD chaining, each associated with appropriate documentation.

Any expansion of the TEI beyond its traditional user base involves a recognition that there are many differing answers to the traditional question "What is text, really?" We report on some work carried out in the context of the COST Action Distant Reading for European Literary History (CA16204), in particular on the TEI-conformant schemas developed for one of its principal deliverables: the European Literary Text Collection (ELTeC). The ELTeC will contain comparable corpora for each of at least a dozen European languages, each being a balanced sample of one hundred novels from the period 1840 to 1920, together with metadata concerning their production and reception. We hope that it will become a reliable basis for comparative work in data-driven textual analytics. The focus of the ELTeC encoding scheme is not to represent texts in all their original complexity, nor to duplicate the work of scholarly editors. Instead, we aim to facilitate a richer and betterinformed distant reading than a transcription of lexical content alone would permit. At the same time, where the TEI encourages diversity, we enforce consistency by permitting representation of only a specific and quite small set of textual features, both structural and analytical. These constraints are expressed by a master TEI ODD, from which we derive three different schemas by ODD chaining, each associated with appropriate documentation. [ABSTRACT FROM AUTHOR]

A corpus of the Italian local press. This paper introduces CoSIL, a corpus of articles from Italian local newspapers containing about 180,000 texts and 66,000,000 words. The corpus was built to provide researchers with a freely downloadable balanced corpus of journalistic texts and a material for linguistic research on online local press, a nowadays-pervasive source of information. Besides the objectives behind the construction of the corpus, the paper describes its design and development, focusing on its representativeness and balance.

This study explores the influence of corpus design when comparing lexical bundle use across groups, examining how the number of texts and average length of texts can impact conclusions about group differences. The study compares the use of lexical bundles by L1-English versus L2-English writers, based on analysis of two sub-corpora of academic articles that are matched for discipline, writer expertize, time of publication, and audience. However, the two sub-corpora differ with respect to the number of texts and the average length of texts. Three experiments examined the influence of differences in corpus composition. The results show that differences in the number of words and number of texts across sub-corpora can have a strong effect on claimed differences in bundle use across groups. This effect is found even when the texts in the corpora are closely matched for their register and topic. [ABSTRACT FROM AUTHOR]

This paper reports on the construction of the Sydney Corpus of Television Dialogue (SydTV). SydTV comprises approximately 275,000-words of dialogue from sixty-six episodes of recent US American fictional television series. The paper first provides a brief overview of existing TV dialogue corpora and then outlines the basic corpus composition, the corpus design principles, and the processes of data collection and storage. SydTV is a small, specialised corpus designed with the objective of being representative of fictional US TV dialogue. TV dialogue is defined as the dialogue uttered by actors on screen as they are performing characters in fictional TV series. The corpus is fairly balanced, since it contains 116,295 words from drama genres and 158,779 words from comedy genres as well as 135,887 words from 'quality' and 139,187 words from 'mainstream' TV series, in addition to a healthy mix of different types of episodes in terms of textual time (pilot episodes, final episodes, episodes occurring at the beginning, middle or end of a season). The corpus is available for educational (teaching and research) purposes through an online interface and has a companion website where frequency lists are provided. [ABSTRACT FROM AUTHOR]

Lexical dispersion is typically measured across arbitrary corpus parts of equal size. In this study, we apply DA – a new dispersion index designed for unequal-sized corpus parts – to the British National Corpus (BNC) in a series of cases studies to show that the dispersion of a word is strongly influenced by the corpus units or parts it is measured across. Our results show that dispersion should be measured and interpreted based on corpus units that are linguistically meaningful for a particular research goal. We conclude with recommendations to help researchers select meaningful corpus units for measuring and interpreting lexical dispersion. [ABSTRACT FROM AUTHOR]

DIALEKT, a corpus of Czech dialects, has been continuously curated and expanded by the Spoken Corpora section of the Institute of the Czech National Corpus. The following paper aims first to give a concise characteristic of the corpus, addressing its sociolinguistic parameters and possible subcorpora derivable thereof, its two-layer approach to the transcription of dialect recordings, and lemmatization & morphological tagging of the corpus. Subsequently, we move on to examples of how linguists can use the corpus and discuss two related projects which expand upon currently available possibilities: an archive of dialect-specific differential phones of the Czech language (completed) and an interactive web environment for spatial map-based visualization of data from all kinds of spoken corpora (in preparation). Thanks in part also to these additional tools, the DIALEKT corpus should serve both experts in the field as well as the general public. [ABSTRACT FROM AUTHOR]

We present JNV (Japanese Nonverbal Vocalizations) corpus, a corpus of Japanese nonverbal vocalizations (NVs) with diverse phrases and emotions. Existing Japanese NV corpora either lack phrase diversity or focus on a small number of emotions, which makes it difficult to analyze the characteristics of Japanese NVs and support downstream tasks like emotion recognition. We first propose a corpus-design method that contains two phases: (1) collecting NVs phrases based on crowd-sourcing; (2) recording NVs by stimulating speakers with emotional scenarios. We then collect 420 audio clips from 4 speakers that cover 6 emotions based on the proposed method. Results of comprehensive objective and subjective experiments demonstrate that (1) the emotions of the collected NVs can be recognized with high accuracy by both human evaluators and statistical models; (2) the collected NVs have a high authenticity comparable to previous corpora of English NVs. Additionally, we analyze the distributions of vowel types in Japanese and conduct feature importance analysis to show discriminative acoustic features between emotion categories in Japanese NVs. We publicate JNV to advance further development in this field. • A Japanese corpus of nonverbal vocalizations called JNV is proposed. • JNV is the largest Japanese nonverbal corpus regarding phrase and emotion diversities. • A scalable, two-phase corpus-design method for nonverbal vocalizations is proposed. • JNV has comparable emotion recognizability and authenticity to English corpora. • Japanese and English frequently use the vowel "a", but "e" only appears in Japanese. [ABSTRACT FROM AUTHOR]

Cognitive linguistics makes specific claims about language use, and corpora are our most powerful tool to test those claims. Representative sampling (Laplace 1814) is a technique that allows us to study smaller, more manageable corpora, and generalize our results to a broader sampling frame. For a sampled corpus to be relevant to our research questions, its sampling frame must have an understandable connection to the subject of our research question.In my dissertation study (Grieve-Smith 2009) I tested the type frequency hypothesis of analogical extension (Bybee 1995) using the FRANTEXT corpus (CNRTL 2018). In this study I test the theatrical texts in FRANTEXT from 1800-1815 against the new Digital Parisian Stage corpus, sampled from Wicks (1950 et seq.), a catalog of every play that premiered in Paris in the nineteenth century. Declarative sentence negations in the Digital Parisian Stage corpus occurred with ne … pas in 73.9 % of tokens, while in FRANTEXT they only occurred with ne … pas in 50.5 % of tokens. This shows that FRANTEXT is biased in favor of elite literary language. To properly test usage-based theories of language change we will need a representative corpus covering a century or more.

This paper introduces a new corpus resource for language learning research, the Trinity Lancaster Corpus (TLC), which contains 4.2 million words of interaction between L1 and L2 speakers of English. The corpus includes spoken production from over 2,000 L2 speakers from different linguistic and cultural backgrounds at different levels of proficiency engaged in two to four tasks. The paper provides a description of the TLC and places it in the context of current learner corpus development and research. The discussion of practical decisions taken in the construction of the TLC also enables a critical reflection on current methodological issues in corpus construction. [ABSTRACT FROM AUTHOR]

Recently, more attention has been paid to the issues of corpus design and representativeness. These issues are especially important for general-purpose language corpora such as the spoken corpora developed within the framework of the Czech National Corpus. This text is a response to Jan Chromý's paper "Comparison of spoken corpora from a sociolinguistic perspective" (Slovo a slovesnost 78, 2017: 145-158), in which the author compares the general-purpose spoken corpus ORAL2013 with his own dataset collected for the SAUP project. We argue that some of his claims are not justified by the findings presented in the paper and that his understanding of the concept of representativeness is rather misleading. Therefore, we aim to clarify some fundamental design decisions adopted for the compilation of ORAL2013 by responding to the specific objections raised by Chromý. We also point out some methodological and reasoning inconsistencies in his paper. [ABSTRACT FROM AUTHOR]

The paper presents a corpus of spontaneous spoken Czech called ORAL2013, its design principles and practical solutions adopted during the data collection. The corpus is designed to represent contemporaryspontaneous spoken language used in informal, real-life situations across the whole of the Czech Republic. The corpus consists of audio recordings and their transcriptions aligned with time stamps; it features manual annotation and broad regional coverage with a large variety of speakers. ORAL2013 contains 835 recordings from the period 2008 to 2011 made with 2,544 speakers (of whom 1,297 speakers are unique); the total length of the audio tracks is almost 300 hours and the total size of the transcriptions exceeds 3.28 million tokens. ORAL2013 is made publicly available by the Czech National Corpus at http://www.korpus.cz/.