Open Corpus Research Articles

Annotated Open Corpus Construction and BERT-Based Approach for Automatic Metadata Extraction From Korean Academic Papers

With the accelerating development of science and technology, the academic papers being published in various fields are increasing rapidly. Academic papers specially in science and technology fields are a crucial media for researchers who develop new technologies by identifying knowledge regarding the latest technological trends and conduct derivative studies in science and technology. Therefore, the continual collection of extensive academic papers, structuring of metadata, and construction of databases are significant tasks. However, research on automatic metadata extraction from Korean papers is not being actively conducted currently owing to insufficient Korean training data. We automatically constructed the largest labeled corpus in South Korea to date from 315,320 PDF papers belonging to 503 Korean academic journals and this labeled corpus can be used for training the models of automatic extraction for 12 metadata types from PDF papers. This labeled corpus is available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://doi.org/10.23057/48</uri> . Moreover, we developed inspection process and guidelines for the automatically constructed data and performed a full inspection of the validation and testing data. The reliability of the inspected data was verified through the inter-annotator agreement measurement. Using our corpus, we trained and evaluated the BERT based transfer learning model to verify its reliability. Furthermore, we proposed new training methods that can improve the metadata extraction performance of Korean papers, and through these methods, we developed KorSciBERT-ME-J and KorSciBERT-ME-J+C models. The KorSciBERT-ME-J showed the highest performance with an F1 score of 99.36%, as well as robust performance in automatic metadata extraction from Korean academic papers in various formats.

Criteria for sample sentences in phraseological dialect dictionaries: a proposal based onGEPHRAS2

AbstractThis article offers a preliminary overview of dialect phraseography, showing that it is still a relatively new field of study. However, as far as Genoese-Italian phraseography is concerned, significant developments have been made in recent years as a result of theGEPHRASandGEPHRAS2projects.The two projects (P31321 and P 33303-G) have been financed by the Austrian Science Fund FWF and led by E. Autelli. Many thanks to Christina Scharf and Anne Freckleton for proofreading this article.The second part of this article focuses on sample sentences found in dialect dictionaries, while the third part will briefly explain theGEPHRAS2project. This will be followed by a proposal as to how to create sample sentences for dialect phraseological dictionaries according to the criteria chosen forGEPHRAS2, and how to analyse ancient and modern phrasemes. It will be shown that one cannot always rely on corpora in order to document sample sentences in dictionaries because, for some dialects, open access corpus is not always available. It will also be seen that, in the preparation of sample sentences, it is essential to establish a set of criteria in order to guarantee readability and authenticity, and to offer concrete help to dictionary users.

Open Chinese Internet Sarcasm Corpus Construction: An Approach

Sarcasm is a commonly-used language phenomenon particularly on the Internet, which is often to convey criticism or negative emotions. A proper sarcasm corpus to help sarcasm study and detection can contribute to linguistic research and assist sentiment analysis, but an open Chinese corpus is found extremely lacking. In this paper, we referenced existing methods and data and constructed a balanced open Chinese Internet sarcasm corpus in a new approach to improve efficiency and data quality. The balanced open corpus contains multi-source and labeled 2,000 texts selected from bigger corresponding origin datasets. In our corpus, sarcasm and non-sarcasm, longer and shorter texts are both in 1:1 ratio.

Speech acts in the Dutch COVID-19 Press Conferences

An open source corpus of all Dutch COVID-19 Press Conferences with sentences annotated on the basis of John Searle’s Speech Act taxonomy was created. It contains all 58 press conferences held between March 6 2020 and April 20 2021 and has 9.441 manually annotated sentences. Speech acts were annotated in a consistent manner, with a Krippendorff’s alpha of .71. The corpus is easy to use and rich in metadata, with lexical, syntactic, discourse (speaker, question or answer) features and information on the type of regulations being present. We analyse the press conferences in terms of speech act usage, giving insight into the use of speech acts over time, the relation of speech act usage to real world phenomena, the general structure of the press conferences and the division of roles between speakers. Relations were found between speech act usage and the type of press conference (i.e. easing, tightening or neutral) as well as the number of hospital admissions. Speech act classes showed preferred locations within the press conferences, indicating a general structure. Distinct roles between speakers were identified. We also investigate the use of our set of labelled sentences for training a speech act classifier and achieve a reasonable accuracy of .73 and a mean reciprocal rank of .74 with the state of the art transformer RoBERTa model.

Recursive Sentiment Detection Algorithm for Russian Sentences

The article is devoted to the task of sentiment detection of Russian sentences. The sentiment is conceived as the author's attitude to the topic of a sentence. This assay considers positive, neutral, and negative sentiment classes, i.e., the task of three-classes classification is solved. The article introduces a rule-based sentiment detection algorithm for Russian sentences. The algorithm is based on the assumption that the sentiment of a phrase can be determined by the sentiments of its parts by the recursive application of appropriate semantic rules to the sentiments of its parts organized as a constituency parse tree. The utilized set of semantic rules was constructed based on a discussion with experts in linguistics. The experiments showed that the proposed recursive algorithm performs slightly worse on the hotel reviews corpus than the adapted rule-based approach: weighted $F_1$-measures are 0.75 and 0.78, respectively. To measure the algorithm efficiency on complex sentences, we created OpenSentimentCorpus based on OpenCorpora, an open corpus of sentences extracted from Russian news and periodicals. On OpenSentimentCorpus the recursive algorithm performs better than the adapted approach does: $F_1$-measures are 0.70 and 0.63, respectively. This indicates that the proposed algorithm has an advantage in case of more complex sentences with more subtle ways of expressing the sentiment.

Dataset Augmentation Using Back-Translation to Improve Early Stage Dialog Systems

As dialog systems are increasingly used, a major challenge for building new ones is the lack of annotated training data. The necessary data collection and annotation efforts are laborious and time-consuming. A potential solution is to augment initial seed data by automatically paraphrasing existing samples. In this paper, we propose a novel data-efficient approach towards this goal. Our method can kick-start a dialog system with minimum human effort while delivering a performance strong enough to allow real-world usage. We ran experiments using Neural Machine Translation on two open corpora. On both of them, the proposed approach improved the generalization capabilities of the model. Our results suggest that paraphrase generation techniques could be used as-is to provide a boost in performance to dialog systems in an early phase.

The Open Corpus of the Veps and Karelian Languages: Overview and Applications

Abstract. A growing priority in the study of Baltic-Finnic languages of the Republic of Karelia has been the methods and tools of corpus linguistics. Since 2016, linguists, mathematicians, and programmers at the Karelian Research Centre have been working with the Open Corpus of the Veps and Karelian Languages (VepKar), which is an extension of the Veps Corpus created in 2009. The VepKar corpus comprises texts in Karelian and Veps, multifunctional dictionaries linked to them, and software with an advanced system of search using various criteria of the texts (language, genre, etc.) and numerous linguistic categories (lexical and grammatical search in texts was implemented thanks to the generator of word forms that we created earlier). A corpus of 3000 texts was compiled, texts were uploaded and marked up, the system for classifying texts into languages, dialects, types and genres was introduced, and the word-form generator was created. Future plans include developing a speech module for working with audio recordings and a syntactic tagging module using morphological analysis outputs. Owing to continuous functional advancements in the corpus manager and ongoing VepKar enrichment with new material and text markup, users can handle a wide range of scientific and applied tasks. In creating the universal national VepKar corpus, its developers and managers strive to preserve and exhibit as fully as possible the state of the Veps and Karelian languages in the 19

Age group classification and gender recognition from speech with temporal convolutional neural networks

This paper analyses the performance of different types of Deep Neural Networks to jointly estimate age and identify gender from speech, to be applied in Interactive Voice Response systems available in call centres. Deep Neural Networks are used, because they have recently demonstrated discriminative and representation capabilities in a wide range of applications, including speech processing problems based on feature extraction and selection. Networks with different sizes are analysed to obtain information on how performance depends on the network architecture and the number of free parameters. The speech corpus used for the experiments is Mozilla’s Common Voice dataset, an open and crowdsourced speech corpus. The results are really good for gender classification, independently of the type of neural network, but improve with the network size. Regarding the classification by age groups, the combination of convolutional neural networks and temporal neural networks seems to be the best option among the analysed, and again, the larger the size of the network, the better the results. The results are promising for use in IVR systems, with the best systems achieving a gender identification error of less than 2% and a classification error by age group of less than 20%.

A Novel Machine Lip Reading Model

Lip Reading is the technology of obtaining the language content by analyzing the change of the speaker’s lip shape and recognizing the information of the lip movement. Lip reading helps people with hearing disabilities understand what other people are saying, which is difficult for humans. This paper proposes a novel Lip Reading model using Transformer network, to achieve a lip reading model with high accuracy. The main process of the model includes the processing of data sets, the extraction of lip features using the pre-trained neural network, and then input into Transformer network for training. Finally, our model achieves a word-level lip reading accuracy of 45.81% on the open source GRID corpus.

Building A Corpus of Open Ancient Greek: New Ways to Learn, Teach, and Publish

Documents the history and rationale as well as the technology for the creation of an open online corpus of Ancient Greek texts, the Free First Thousand Years of Greek. Reveals the energy on a worldwide basis for building open data along with the boundless creativity around developing software tools for teaching, research, and publishing that exploit open data in ways that were impossible before.

The protein-protein interaction ontology: for better representing and capturing the biological context of protein interaction

BackgroundWith the rapid increase in the amount of Protein-Protein Interaction (PPI) data, the establishment of an event-centered PPI ontology that contains temporal and spatial vocabularies is urgently needed to clarify PPI biological annotations. In this paper, we propose a precisely designed schema - PPIO (PPI Ontology) for representing the biological context of PPIs.ResultsInspired by the event model and the distinct characteristics of PPI events, PPIO consists of six core aspects of the information required for reporting a PPI event, including the interactor (who), the biological process (when), the subcellular location (where), the interaction type (how), the biological function (what) and the detection method (which). PPIO is implemented through the integration of appropriate terms from the corresponding vocabularies/ontologies, e.g., Gene Ontology, Protein Ontology, PSI-MI/MOD, etc. To assess PPIO, an approach based on PPIO in developed to extract PPI biological annotations from an open standard corpus “BioCreAtIvE-PPI”. The experiment results demonstrate PPIO’s high performance, a precision of 0.69, a recall of 0.72 and an F-score of 0.70.ConclusionsPPIO is a well-constructed essential ontology in the interpretation of PPI biological context. The results of the experiments conducted on the BioCreAtIvE corpus demonstrate that PPIO is able to facilitate PPI annotation extraction from biomedical literature effectively and enrich essential annotation for PPIs.

TalkBank의 음성 말뭉치를 활용한 대화 참여자의 상호작용 연구

This paper analyzes speaker dynamics in conversational exchanges between mothers and their young children based on open corpus data in the TalkBank system. In particular, we analyze four mother-infant dyads in the Providence corpus with focus on the interaction of the initiator of Conversational Block and the Conversational Turn with regard to variables such as the turn length and the response time. We show that both the mother and child speak longer in conversational blocks that they have initiated, which is consistent with the findings in Ko et al. (2016) based on data collected and processed by the LENA^® system. The model for response time, however, did not yield a significant interaction despite some indications towards such a tendency. The methodology for conducting such analyses using the corpus data in TalkBank is described in detail so that it could serve as a reference for research adopting similar procedures.

THE LINGUISTIC CORPUS VEPKAR IS A LANGUAGE REFUGE FOR THE BALTICFINNISH LANGUAGES OF KARELIA

The purpose of creating conservation areas is to protect endangered plant and animal species. Large, tagged linguistic corpora with a great variety of genres are used for the preservation and research of safe and endangered languages. The article describes the history, structure and development of the Open Corpus of the Veps and Karelian languages. The Veps language corpus was created in 2009 under the leadership of Nina Zaitseva. Three Karelian subcorpora (Karelian proper, Livvi and Ludian) were included in the linguistic corpus in 2016. The united linguistic platform was named “The Open Corpus of the Veps and Karelian languages” (VepKar). This linguistic corpus includes texts and dictionaries stored in a database, and a computer program (corpus manager) for searching and processing the data. This corpus manager was written in the PHP programming language in the Laravel framework. The data are stored in a MySQL database. Corpus and dictionaries data are available online (dictorpus.krc.karelia.ru). YouTube and Wikipedia are used by VepKar authors to popularize the corpus. Dictionaries and corpus texts are strongly interrelated. Multifunctional dictionaries of the Veps and Karelian languages contain definition, translation, dialect labels, semantic relations (synonyms, antonyms, etc.), examples of word usage with reference to texts, as well as complete inflectional paradigms. All texts are automatically marked up and there are references from words in the text to the corresponding meanings in the dictionary entries. The developers continue adding useful new features to the corpus manager to make the work of editors easier. For example, over the past three years, nominal and verbal inflection rules have been formulated and programmed for all dialects of the Veps language and its newly-written version, as well as for the Livvi-Karelian, North Karelian and Tver newly-written versions of the Karelian language. Thanks to this, 2.1 million word forms were generated in the VepKar system in a semi-automatic mode. The semantic markup in the corpus is 2.1 million links between words from the text and the meanings of lemmas in the dictionary. The grammatical markup was added, namely, 1.1 million links between words from the text and the grammatical features of word forms from the dictionary were automatically established. The multilingual VepKar corpus is divided into subcorpora according to languages and dialects, and the texts are also classified into styles and genres. The corpus has a sophisticated search system (with filtering of texts by language, style and dialect, by informant, collector or author, by year of recording or year of publication). It is possible to search for lemmas by dialects, parts of speech, grammatical features, and even by lexical-semantic categories. These categories appeared due to the integration of the data of the outstanding “Comparative and Onomasiological Dictionary of the Dialects of the Karelian, Veps and Sami Languages” into the vocabulary part of VepKar. In 2021, the Sanahelmi electronic dictionary was created on the basis of VepKar for Android phones. The development of mobile applications based on corpus data is our bright future.

Gender Bias in the News: A Scalable Topic Modelling and Visualization Framework.

We present a topic modelling and data visualization methodology to examine gender-based disparities in news articles by topic. Existing research in topic modelling is largely focused on the text mining of closed corpora, i.e., those that include a fixed collection of composite texts. We showcase a methodology to discover topics via Latent Dirichlet Allocation, which can reliably produce human-interpretable topics over an open news corpus that continually grows with time. Our system generates topics, or distributions of keywords, for news articles on a monthly basis, to consistently detect key events and trends aligned with events in the real world. Findings from 2 years worth of news articles in mainstream English-language Canadian media indicate that certain topics feature either women or men more prominently and exhibit different types of language. Perhaps unsurprisingly, topics such as lifestyle, entertainment, and healthcare tend to be prominent in articles that quote more women than men. Topics such as sports, politics, and business are characteristic of articles that quote more men than women. The data shows a self-reinforcing gendered division of duties and representation in society. Quoting female sources more frequently in a caregiving role and quoting male sources more frequently in political and business roles enshrines women’s status as caregivers and men’s status as leaders and breadwinners. Our results can help journalists and policy makers better understand the unequal gender representation of those quoted in the news and facilitate news organizations’ efforts to achieve gender parity in their sources. The proposed methodology is robust, reproducible, and scalable to very large corpora, and can be used for similar studies involving unsupervised topic modelling and language analyses.

Retrospective analysis of open surgical versus laser interstitial thermal therapy callosotomy in pediatric patients with refractory epilepsy.

Corpus callosotomy remains an established surgical treatment for certain types of medically refractory epilepsy in pediatric patients. While the traditional surgical approach is often well tolerated, the advent of MR-guided laser interstitial thermal therapy (LITT) provides a new opportunity to ablate the callosal body in a minimally invasive fashion and minimize the risks associated with an open interhemispheric approach. However, the literature is sparse regarding the comparative efficacy and safety profiles of open corpus callosotomy (OCC) and LITT callosotomy. To this end, the authors present a novel retrospective analysis comparing the efficacy and safety of these methods. Patients who underwent OCC and LITT callosotomy during the period from 2005 to 2018 were included in a single-center retrospective analysis. Patient demographic and procedural variables were collected, including length of stay, procedural blood loss, corticosteroid requirements, postsurgical complications, and postoperative disposition. Pre- and postoperative seizure frequency (according to seizure type) were recorded. In total, 19 patients, who underwent 24 interventions (16 OCC and 8 LITT), were included in the analysis. The mean follow-up durations for the OCC and LITT cohorts were 83.5 months and 12.3 months, respectively. Both groups experienced reduced frequencies of seizure and drop attack frequency postoperatively. Additionally, LITT callosotomy was associated with a significant decrease in estimated blood loss and decreased length of pediatric ICU stay, with a trend of shorter length of hospitalization. Longer-term follow-up and a larger population are required to further delineate the comparative efficacies of LITT callosotomy and OCC for the treatment of pediatric medically refractory epilepsy. However, the authors' data demonstrate that LITT shows promise as a safe and effective alternative to OCC.

The noun visilka: a semantic portrait and a sys­tem of multiple meanings (based on police search documentation of the second half of the 19th century)

his article is devoted to one of the important aspects of studying the Russian business language of the second half of the 19th century — the formation of the lexical meaning of the noun visilka (exile, expulsion), typical of the police procedure documentation of the time. The author discovered business texts of this type in the funds of the State Archive of the Tyumen region, the State Archive of the Omsk region, and the State Archive of the Irkutsk region. All texts date back to the late 19th century. The study of the lexeme visilka was conducted in ac­cordance with the lexicographic portraiture methodology (proposed by Apresyan and Mayorov) and the principles of linguistic study of source texts. A comprehensive analysis of the lexical distribution of the noun visilka determined the necessity of abstract or concrete-abstract vector locatives, which expanded the lexical meaning of this lexeme in the business texts of the second half of the 19th century. The authors identified the main semantic features of the noun visilka (the dominance of the archiseme — 'process', the existence of the differen­tial semes 'result', 'actions of people involved in a police search'). Further research is aimed at examining business writings from Western and Eastern Siberia, which belong to a different genre and a different time period. Further research will be based on the texts from the Nation­al Corpus of the Russian language, the Uppsala and Tyubinsk corpuses, the database of the Russian press Integrum and Open Corpus.

Are You Being Rhetorical? A Description of Rhetorical Move Annotation Tools and Open Corpus of Sample Machine-Annotated Rhetorical Moves

Writing analytics has emerged as a sub-field of learning analytics, with applications including the provision of formative feedback to students in developing their writing capacities. Rhetorical markers in writing have become a key feature in this feedback, with a number of tools being developed across research and teaching contexts. However, there is no shared corpus of texts annotated by these tools, nor is it clear how the tool annotations compare. Thus, resources are scarce for comparing tools for both tool development and pedagogic purposes. In this paper, we conduct such a comparison and introduce a sample corpus of texts representative of the particular genres, a subset of which has been annotated using three rhetorical analysis tools (one of which has two versions). This paper aims to provide both a description of the tools and a shared dataset in order to support extensions of existing analyses and tool design in support of writing skill development. We intend the description of these tools, which share a focus on rhetorical structures, alongside the corpus, to be a preliminary step to enable further research, with regard to both tool development and tool interaction

An Empirical Study on Deep Neural Network Models for Chinese Dialogue Generation

The task of dialogue generation has attracted increasing attention due to its diverse downstream applications, such as question-answering systems and chatbots. Recently, the deep neural network (DNN)-based dialogue generation models have achieved superior performance against conventional models utilizing statistical machine learning methods. However, despite that an enormous number of state-of-the-art DNN-based models have been proposed, there lacks detailed empirical comparative analysis for them on the open Chinese corpus. As a result, relevant researchers and engineers might find it hard to get an intuitive understanding of the current research progress. To address this challenge, we conducted an empirical study for state-of-the-art DNN-based dialogue generation models in various Chinese corpora. Specifically, extensive experiments were performed on several well-known single-turn and multi-turn dialogue corpora, including KdConv, Weibo, and Douban, to evaluate a wide range of dialogue generation models that are based on the symmetrical architecture of Seq2Seq, RNNSearch, transformer, generative adversarial nets, and reinforcement learning respectively. Moreover, we paid special attention to the prevalent pre-trained model for the quality of dialogue generation. Their performances were evaluated by four widely-used metrics in this area: BLEU, pseudo, distinct, and rouge. Finally, we report a case study to show example responses generated by these models separately.

Cross-Domain Tibetan Word Segmentation Based on Deep Learning

Tibetan word segmentation is the primary task of various types of Tibetan natural language processing. However, when the test corpus and the training corpus are inconsistent, the accuracy of word segmentation will drop significantly. Therefore, in this paper firstly we build a BiLSTM-CRF Tibetan word segmentation model based on deep learning that includes dictionary features, and train the model parameters on the open domain word segmentation data. Then, taking the Tibetan medical clinical text as a small-scale word segmentation training corpus, and fine-tuning the parameters of the BiLSTM-CRF word segmentation model of the open domain corpus, and adding the domain vocabulary to the dictionary features of the model. So that the precision, recall rate and F value of the word segmentation model on the test data reached 93.21%, 92.10%, and 94.37% respectively.

A Unified Framework for Multilingual Speech Recognition in Air Traffic Control Systems

This work focuses on robust speech recognition in air traffic control (ATC) by designing a novel processing paradigm to integrate multilingual speech recognition into a single framework using three cascaded modules: an acoustic model (AM), a pronunciation model (PM), and a language model (LM). The AM converts ATC speech into phoneme-based text sequences that the PM then translates into a word-based sequence, which is the ultimate goal of this research. The LM corrects both phoneme- and word-based errors in the decoding results. The AM, including the convolutional neural network (CNN) and recurrent neural network (RNN), considers the spatial and temporal dependences of the speech features and is trained by the connectionist temporal classification loss. To cope with radio transmission noise and diversity among speakers, a multiscale CNN architecture is proposed to fit the diverse data distributions and improve the performance. Phoneme-to-word translation is addressed via a proposed machine translation PM with an encoder-decoder architecture. RNN-based LMs are trained to consider the code-switching specificity of the ATC speech by building dependences with common words. We validate the proposed approach using large amounts of real Chinese and English ATC recordings and achieve a 3.95% label error rate on Chinese characters and English words, outperforming other popular approaches. The decoding efficiency is also comparable to that of the end-to-end model, and its generalizability is validated on several open corpora, making it suitable for real-time approaches to further support ATC applications, such as ATC prediction and safety checking.