Language Corpora Research Articles

Электронный корпус татарского языка на базе модели лингвистических графов знаний

The article presents a new version of the electronic corpus of the Tatar language, updated based on a linguistic knowledge graph model for Turkic languages. This new version of the corpus allows for information description across multiple linguistic levels: morphonological, syntactic, and semantic, through the use of knowledge graphs to represent linguistic data. This approach enhances corpus functionality, enabling searches that incorporate syntactic and semantic information. A distinctive feature of the electronic corpus implementation is that the model employed aligns closely with the structural and functional characteristics of Turkic languages and serves as a foundation for developing various software products for semantic text processing in Turkic languages. In particular, these products include the linguistic portal "Turkic Morphme" and the new version of the Tatar language electronic corpus, "Tugan Tel.".

Russian academic lexicography and the National dictionary foundation

The article is devoted to modern trends in the development of Russian linguistics, primarily to dictionaries and the National Corpus of the Russian Language. It is noted that the corpus representation of a language does not exist autonomously; it influences traditional forms of linguistic work and makes it possible to rebuild the technology of lexicography. Nowadays, a printed dictionary can no longer provide complete information in a time-appropriate form. Therefore, many countries are currently creating various systems for electronic presentation of national dictionaries and connecting them with the language corpus. Such work is actively carried out in Russia, primarily by V.V. Vinogradov Institute of Russian Language of the RAS.

The open corpus of the Vepsian and Karelian languages

In order to preserve and systematically study the Vepsian and Karelian languages, the staff of the Institute of Language, Literature and History and the Institute of Applied Mathematical Research of the KarSC RAS have created and continue to improve the VepKar language corpus. The project is aimed at preserving and accumulating written texts in Karelian and Vepsian languages, fixing and preserving oral speech, researching the Baltic-Finnish languages of Karelia, editing the norms of newly written versions of Karelian and Vepsian languages, and creating educational applications. Anyone can use VepKar as an electronic library and a full-fledged electronic dictionary, which makes this resource very popular.

Beyond breathalyzers: AI-powered speech analysis for alcohol intoxication detection

Detecting potential alcohol inebriation or intoxication status holds paramount significance for social prevention and security. Beyond its association with long-term health effects, alcohol consumption can lead to immediate consequences, including reduced control over one’s actions, with traffic fatalities representing one of the most tragic outcomes.This study leveraged the Alcohol Language corpus, involving 162 subjects recorded both in sober and inebriated states. Participants provided 60 speech samples while sober and 30 when intoxicated, all within a realistic car setting using head-mounted microphones. Our research endeavors encompassed comprehensive stratified statistical tests to examine the impact of alcohol consumption on speech production while uncovering the influence of covariates such as age, gender, and drinking habits.Additionally, we introduced a speaker-neutral machine learning algorithm, based on the Domain-Adversarial Neural Network architecture. This approach aimed to overcome challenges posed by individual differences that often complicate intoxicated speech analysis. Notably, our findings highlighted the effectiveness of features like the RASTA-filtered auditory spectrum. Nevertheless, the results from statistical tests emphasized the need for techniques that minimize inter-subject variability.As for the automatic classification, the proposed architecture exhibited promising results, yielding a classification accuracy slightly exceeding 70% on an independent test set. Although preliminary, our research demonstrates the potential for detecting alcohol-induced speech changes, benefiting societal well-being and security. It also underscores the importance of developing strategies that account for individual differences while harnessing the power of automatic models to effectively distinguish between sober and intoxicated individuals.

Milestones in Polish: From roadside signs to conditions (not) to be met

The article presents the history of the lexical unit kamień milowy ‘milestone’ in the Polish language, with a particular emphasis on its new meaning, coming from the EU variety of English, and the meaning-specific collocability. The study has shown how the political context influenced the popularity of the name in its new meaning, its functioning in the media, and its social reception. Data come from dictionaries, language corpora, the Polona digital library, the EU legal acts database, and the EU terminology database

A large corpus for the recognition of Greek Sign Language gestures

Sign language recognition (SLR) from videos constitutes a captivating problem in gesture recognition, requiring the interpretation of hand movements, facial expressions, and body postures. The complexity of sign formation, signing variability among signers, and the technical hurdles of visual detection and tracking render SLR a challenging task. At the same time, the scarcity of large-scale SLR datasets, which are critical for developing robust data-intensive deep-learning SLR models, exacerbates these issues. In this article, we introduce a multi-signer video corpus of Greek Sign Language (GSL), which is the largest GSL database to date, serving as a valuable resource for SLR research. This corpus comprises an extensive RGB+D video collection that conveys rich lexical content in a multi-modal fashion, encompassing three subsets: (i) isolated signs; (ii) continuous signing; and (iii) continuous alphabet fingerspelling of words. Moreover, we introduce a comprehensive experimental setup that paves the way for more accurate and robust SLR solutions. In particular, except for the multi-signer (MS) and signer-independent (SI) settings, we employ a signer-adapted (SA) experimental paradigm, facilitating a comprehensive evaluation of system performance across various scenarios. Further, we provide three baseline SLR systems for isolated signs, continuous signing, and continuous fingerspelling. These systems leverage cutting-edge methods in deep learning and sequence modeling to capture the intricate temporal dynamics inherent in sign gestures. The models are evaluated on the three corpus subsets, setting their state-of-the-art recognition benchmark. The SL-ReDu GSL corpus, including its recommended experimental frameworks, is publicly available at https://sl-redu.e-ce.uth.gr/corpus.

Correction to: Advances in sign language corpus linguistics: exploring the frontiers of deaf studies and education.

Correction to: Advances in sign language corpus linguistics: exploring the frontiers of deaf studies and education.

Corpus research on signed languages in the Nordic countries

Abstract This semi-systematic literature review examines signed language corpus research in the Nordic countries, with a quantitative and qualitative assessment of corpus research. The article first describes some critical components and functionalities of signed language corpora. It then outlines the evolution of Nordic corpus research, highlighting Sweden’s pioneering role and subsequent developments in Finland and Norway. The findings suggest a progression from method-focused publications to those exploring linguistic phenomena within and across (signed) languages. Although the number of research publications is modest, there is a discernible shift towards comparative studies and applications in signed language teaching and learning.

Challenges in Intellectualizing Sesotho for Use in Academic Publications

This paper explores challenges in intellectualizing Sesotho for academic works beyond literary-based theses. It also overviews resources for the intellectualization of official Indigenous languages of South Africa, focusing on national translation projects for terminology development. Furthermore, it highlights efforts by the South African Centre for Digital Language Resources in creating and collecting corpora for all official languages, including Sesotho. Challenges in Sesotho lexicography, especially with corpus-based approaches, are discussed. Additionally, the paper reflects on the authors' experiences with publishing academic articles in Sesotho in South African journals. It discusses obstacles faced, and strategies employed, including dialogues with reviewers on creating new terms and choosing between Lesothan Sesotho and South African Sesotho orthographies. Finally, the essay addresses challenges posed by limited publication avenues for indigenous languages in southern Africa and academic pressures for citations and research impact. It aims to shed light on lesser-known Indigenous languages in Southern Africa by spotlighting challenges and solutions to academic publications in Sesotho. Overall, the paper concludes that efforts at intellectualizing Sesotho and other South African indigenous languages are evident. Even so, there is far less emphasis on using Sesotho and other indigenous languages for academic discourses.

Comparative analysis of Russian language textbooks for primary school using quantitative methods

Despite the intense scientific and practical interest in assessing the textual content quality of a school textbook, there is still no transparent algorithm and objective criteria for such evaluation. This article aims to outline objective linguistic criteria for textbook text assessment using quantitative methods on the example of one of the main components of a Russian language textbook, i. e. exercise texts. The analysis was performed on the material of 8 lines of modern Russian language textbooks for primary school and 1 line of the digital educational service Yandex Textbook. The research employed the methods of automatic natural language processing, statistics, and corpus analysis of texts. As a result, we propose an algorithm for analysing the text library of a Russian language textbook according to the criteria of structural and lexical complexity, genre, stylistic and thematic diversity, and the degree of correspondence between the text library and the current state of the Russian language. The obtained results form an idea of the textual content of a modern Russian language textbook for primary school. Moreover, they highlight the specific features of text selection for modern traditional and digital Russian language textbooks and resources. Finally, the research outcomes raise questions concerning the methodological optimality of certain authorial decisions when compiling the text library of a textbook. The results of the study will be useful to authors of school textbooks on the Russian language and specialists in educational and methodological work.

Big Data and Ideational Institutionalism. Reconsidering the Possibilities and Limitations of Google Ngram in the Study of Ideas

AbstractIdeas and ideational change are difficult to grasp. When exactly did the idea of sustainability emerge? How has the meaning of democracy changed over time? Did ideational change precede or follow institutional change? These and many similar questions are highly relevant to the emerging field of ideational institutionalism, but they are difficult to answer conclusively without reliable data. Google Books Ngram Viewer, a tool introduced in 2011, opens up new possibilities in this regard by allowing the analysis of millions of print sources over time. However, after initial euphoria, its use has waned somewhat due to growing doubts about its reliability and the validity of the results it produces. Some of these problems have been addressed, but the debate about the potential and limitations of Google Books Ngram Viewer is ongoing. This research note will contribute to this debate by proposing a systematization for the use of search terms based on the concept of belief systems and by presenting the results of a reliability test for the German language corpus.

Investigating Language Transfer Effect of Arabic and Spanish on English Using Test of English as Foreign Language Corpus

The first language, or also known as a mother tongue, can influence second language acquisition (SLA). Contrastive analysis theory state that mother tongue can either facilitate or hinder second language acquisition. The similarities between first language and second language facilitate the learning process while the differences between L1 and L2 could sometimes hinder the learning process and cause a negative transfer. Thus, to comprehend the impact of L1 on L2, it is necessary to analyze the language performance of EFL learners. The objective of this study was to investigate the negative transfer effect of Arabic and Spanish as a first language (L1) on learning English as a foreign language (L2). The data was collected using test of English as foreign language corpus. The researcher used a qualitative method. The errors were collected and analyzed based on rubric. The findings showed that Arabic and Spanish had a negative transfer effect on English as a second language. Arabic language had more negative transfer effect on English than Spanish language.

SPECIALIZED NEURAL MACHINE TRANSLATION FOR AVIATION ENGLISH: INSIGHTS FROM PORTUGUESE-TO-ENGLISH AVIATION SAFETY REPORTS

Aviation safety reports are essential sources for the identification and analysis of risks in civil aviation. These reports are written in plain language, which requires the application of Natural Language Processing techniques for automatic and intelligent treatment. In the case of Brazil, the vast majority of reports are written in Portuguese. Therefore, for comparison with international database of reports that are written in English, a first step is the translation of Brazilian reports. In this work, a proposal for a machine translation model is presented based on the fine-tuning of pre-trained models. To this end, an aviation-specific language corpus is developed with the objective of generating example data for model training. Finally, a pre-trained model is fine-tuned with the corpus created in order to implement an automatic translation model that achieves good results in the task considering the specifics of the aviation language. As a result, a first model is implemented, presenting coherent results of translation between Portuguese/English in the specific domain of aviation.

How do constructions with modal verbs develop in second language learners of English?

Abstract The study reported on in this paper uses data from a large, pseudolongitudinal corpus of second language (L2) learner writing to investigate how L2 knowledge of verb-argument constructions (VACs) develops from low-beginner to upper-intermediate proficiency levels. The focus is on learners from two L1 backgrounds (L1 German and L1 Spanish) and on a subset of VACs that contain modal auxiliaries in combination with lexical verbs (e.g., subject-modal-verb, as in I can cook). For each VAC, L1, and proficiency level, frequency lists of verbs were created and compared across datasets. Reference data on the same constructions was extracted from a corpus of L1 writing. Results suggest that modal VACs become more frequent, less predictable, more productive, and more aligned with L1 production as learner proficiency increases. The findings of this study help to expand our understanding of the processes that underlie L2 construction acquisition.

GenAI language corpora in foreign language teaching: advantages and disadvantages

Importance. Corpora have proven itself over the years as an effective tool in foreign language teaching. The current stage of technological development is characterized by the introduction of artificial intelligence technologies into all spheres of society. One of the modern programs based on generative artificial intelligence is ChatGPT, which creates answers based on a huge array of data. The processes underlying the functioning of ChatGPT are similar to corpora technologies, which determines the possibility of using ChatGPT as a language corpus. The purpose of the study is to identify the advantages and disadvantages of ChatGPT when used as a language corpus.Research Methods. In the course of the research, the following theoretical methods were used: the study and analysis of scientific literature on the research topic, generalization and classification of conceptual approaches of scientists. As a result of comparing the corpora technologies and the ChatGPT neural network according to nine main criteria, the advantages and disadvantages of each of the programs were identified and described.Results and Discussion. The comparison of corpora technologies and the ChatGPT neural network according to nine main criteria showed that corpora technologies are the preferred choice in mastering lexical and grammatical skills; ChatGPT is preferable when a teacher is in need for quickly prepared training tasks and when teaching students with a lower level of cognitive abilities and foreign language proficiency.Conclusion. ChatGPT neural network can be used in the process of learning a foreign language as a language corpus. However, the application of the program will not always be advisable, and should depend on the training task. The results of the study can be used to develop methods for teaching aspects of language based on corpus-based artificial intelligence technologies.

Using Transfer Learning to Realize Low Resource Dungan Language Speech Synthesis

This article presents a transfer-learning-based method to improve the synthesized speech quality of the low-resource Dungan language. This improvement is accomplished by fine-tuning a pre-trained Mandarin acoustic model to a Dungan language acoustic model using a limited Dungan corpus within the Tacotron2+WaveRNN framework. Our method begins with developing a transformer-based Dungan text analyzer capable of generating unit sequences with embedded prosodic information from Dungan sentences. These unit sequences, along with the speech features, provide <unit sequence with prosodic labels, Mel spectrograms> pairs as the input of Tacotron2 to train the acoustic model. Concurrently, we pre-trained a Tacotron2-based Mandarin acoustic model using a large-scale Mandarin corpus. The model is then fine-tuned with a small-scale Dungan speech corpus to derive a Dungan acoustic model that autonomously learns the alignment and mapping of the units to the spectrograms. The resulting spectrograms are converted into waveforms via the WaveRNN vocoder, facilitating the synthesis of high-quality Mandarin or Dungan speech. Both subjective and objective experiments suggest that the proposed transfer learning-based Dungan speech synthesis achieves superior scores compared to models trained only with the Dungan corpus and other methods. Consequently, our method offers a strategy to achieve speech synthesis for low-resource languages by adding prosodic information and leveraging a similar, high-resource language corpus through transfer learning.

Low Resource Malay Dialect Automatic Speech Recognition Modeling Using Transfer Learning from a Standard Malay Model

Approaches to automatic speech recognition have transited from Hidden Markov Model (HMM)-based ASR to deep neural networks. The advantages of deep neural network approaches are that they can be developed quickly and perform better given large language resources. Nevertheless, dialect speech recognition is still challenging due to the limited resources. Transfer learning approaches have been proposed to improve speech recognition for low resources. In the first approach, the model is pre-trained on a large and diverse labeled dataset to learn the acoustic and language patterns from the speech signal. Then, the model parameters are updated with a new dataset, and the pre-trained model is fine-tuned on a low-resource language dataset. The fine-tuning process is usually completed by freezing the pre-trained layers and training the remaining layers of the model on the low-resource language corpus. Another approach is to use a pre-trained model to capture the compact and meaningful features as input to the encoder. Pre-training in this approach usually involves using unsupervised learning methods to train models on a corpus of large amounts of unmarked data. It enables the model to learn the general patterns and relationships between the input speech signals. This paper proposes a training recipe using transfer learning and Standard Malay models to improve automatic speech recognition for Kelantan and Sarawak Malay dialects.

Prompt response variation in learner corpus research: Implications for data interpretation

While general first language corpora are composed of samples from various naturalistic sources (e.g., websites, books), language samples in most written learner corpora (LC) are texts produced in response to prompts. In this context, LC users need to develop a clear awareness of the affordances and limitations of specific prompts and how responses to said prompts may affect the investigation of their intended object(s) of study. Through an analysis of the presence/absence of specific Spanish verb tenses in texts written in response to two supposedly narrative prompts in a Spanish LC (COWS-L2H; Yamada et al., 2020), this article illustrates the impact of inter- and intra-prompt response variation on LC data interpretation. Based on this evidence, we caution against rapid assumptions about text content based solely on the superficial phrasing of LC writing prompts. Instead, we recommend that LC users perform in-depth quantitative and qualitative analyses of learners’ samples written in response to each prompt they aim to include in their study prior to running statistical models on those data.

Finding continuers in Swedish Sign Language

Abstract In this paper, I use methods from corpus linguistics and computer vision to find candidates for continuers – that is, conversational markers that signal comprehension and encouragement to the primary speaker/signer to continue – in a corpus of Swedish Sign Language (STS). Using different methods based on distributional patterns in conversational turns, I identify a small set of manual signs – particularly the sign JA@ub ‘yes’ – that exhibit the characteristics associated with continuers, such as occurring frequently in repeated sequences of overlapping but noncompetitive turns. The identified signs correspond to those found in previous research on manual backchannels in STS, demonstrating that quantitative, distribution-based approaches are successful in identifying continuers. In a second step, I employ methods from computer vision to analyze a subset of the corpus videos, and find that the continuer candidates show interesting form characteristics: they are small in visible articulation and thus conversationally unobtrusive by often being articulated low and with little movement in signing space. The results show that distribution-based approaches can be used successfully with sign language corpus data, and that the nature of continuers exhibits similarities across modalities of human language.

When the parish turns into a classroom: teaching pluriliteracies in regional collateral languages through place-based education (the case of Latgalian)

ABSTRACT Our paper discusses ways of connecting the acquisition of a regional language with learning processes related to specific places. For this aim, we relate the investigation of the Linguistic Landscape (LL) to the pluriliteracies teaching approach and to place-based learning. As an example, we use the case of Latgalian in Latvia: How can local multilingual and sociocultural realities as identified in the linguistic landscape enrich disciplinary content in regional language teaching, while also enhancing language awareness and active engagement among the students? And how can place-based education be used for developing (pluri)literacy in regional collateral languages? To this end, the paper first provides the background to place-based education, before turning to roles of Latgalian in public space and a recent LL study in Latgale. Then, we provide suggestions of how to ‘turn the parish into a classroom’, but also of how ‘to bring the parish into the class’ by using digital resources, e.g. language corpora. Finally, we suggest a set of components that influence place-based education for developing literacy in a regional collateral language and which may be taken into consideration when preparing educational projects. This paper is complementary to two other papers on Latgalian in this special issue.