General Corpus Research Articles

The influence of the benchmark corpus on keyword analysis

Abstract The growing popularity of keyword analysis as an applied linguistics methodology has not been matched by an increase in the rigour with which the method is applied. While several studies have investigated the impact of choices made at certain stages of the keyword analysis process, the impact of the choice of benchmark corpus has largely been overlooked. In this paper, we compare a target corpus with several benchmark corpora and show that the keywords generated are different. We also show that certain characteristics of the keyword list and of the keywords themselves vary in relatively predictable ways depending on the benchmark corpus. These variations have implications for the choice of benchmark corpus and how the results of a keyword analysis should be interpreted. Analyzing the keywords from a comparison with a large general corpus or the keyword lists from multiple comparisons may be most appropriate for register studies.

DeIDNER Corpus: Annotation of Clinical Discharge Summary Notes for Named Entity Recognition Using BRAT Tool.

Named Entity Recognition (NER) aims to identify and classify entities into predefined categories is a critical pre-processing task in Natural Language Processing (NLP) pipeline. Readily available off-the-shelf NER algorithms or programs are trained on a general corpus and often need to be retrained when applied on a different domain. The end model’s performance depends on the quality of named entities generated by these NER models used in the NLP task. To improve NER model accuracy, researchers build domain-specific corpora for both model training and evaluation. However, in the clinical domain, there is a dearth of training data because of privacy reasons, forcing many studies to use NER models that are trained in the non-clinical domain to generate NER feature-set. Thus, influencing the performance of the downstream NLP tasks like information extraction and de-identification. In this paper, our objective is to create a high quality annotated clinical corpus for training NER models that can be easily generalizable and can be used in a downstream de-identification task to generate named entities feature-set.

Experiência vivenciada por pessoas acometidas por COVID-19 no percurso da internação à alta hospitalar

Objetivo: Analisar a experiência por pessoas acometidas pelo novo coronavírus no percurso da internação à alta. Método: Os dados foram analisados por meio de análises lexicográficas clássicas no Iramuteq. Obteve-se a Classificação Hierárquica Descendente para aferir os dados do dendograma em função das classes geradas, e por fim, a Nuvem de Palavras, que unifica as palavras e dispõem graficamente em função da sua frequência. Resultados: A Classificação Hierárquica Descendente gerou um corpus geral constituído por 17 textos, separados em 79 segmentos de texto, com aproveitamento de 58STs (73,42%). O conteúdo analisado foi categorizado em três classes: Classe 1 (29,31%); Classe 2 (34,48%) e a Classe 3 (36,21%). Conclusão: A experiência vivenciada por pessoas acometidas por COVID-19 no percurso da internação à alta hospitalar na subjetividadee perspectivas de cada ser foi relatada com diferentes impactos físicos, emocionais e sociais que este agente patógeno causa em cada indivíduo.

Collocates of 'great' and 'good' in the Corpus of Contemporary American English and Indonesian EFL textbooks

Previous studies have proven some discrepancies between language presented in EFL textbooks and the real use of English. This fact contradicts the teaching materials’ aim, which is intended to equip the learners to be communicatively competent. This study then investigates the use of adjectives, limited to ‘great’ and ‘good’ in terms of the frequent and strong co-occurrences, i.e., collocations, in a general reference corpus of Standard English to be compared with those in Indonesian EFL textbooks. The data were collected from both the corpus and the textbooks, but the analyses in the textbooks were generated based on the Mutual Information (MI) score of the collocates. Based on the comparison, it is evident that there are some similarities between adjective use in the textbooks and the corpus in terms of verb collocates of the adjectives. The mismatches, however, are quite remarkable, especially in terms of the variability of adverb collocates and the absence of prominent noun collocates in the textbooks. Pedagogically speaking, these results should be taken into consideration in writing the textbooks to enhance the quality of the language content prepared for the learners in the EFL context.

A pre-training and self-training approach for biomedical named entity recognition.

Named entity recognition (NER) is a key component of many scientific literature mining tasks, such as information retrieval, information extraction, and question answering; however, many modern approaches require large amounts of labeled training data in order to be effective. This severely limits the effectiveness of NER models in applications where expert annotations are difficult and expensive to obtain. In this work, we explore the effectiveness of transfer learning and semi-supervised self-training to improve the performance of NER models in biomedical settings with very limited labeled data (250-2000 labeled samples). We first pre-train a BiLSTM-CRF and a BERT model on a very large general biomedical NER corpus such as MedMentions or Semantic Medline, and then we fine-tune the model on a more specific target NER task that has very limited training data; finally, we apply semi-supervised self-training using unlabeled data to further boost model performance. We show that in NER tasks that focus on common biomedical entity types such as those in the Unified Medical Language System (UMLS), combining transfer learning with self-training enables a NER model such as a BiLSTM-CRF or BERT to obtain similar performance with the same model trained on 3x-8x the amount of labeled data. We further show that our approach can also boost performance in a low-resource application where entities types are more rare and not specifically covered in UMLS.

SpaceTransformers: Language Modeling for Space Systems

The transformers architecture and transfer learning have radically modified the Natural Language Processing (NLP) landscape, enabling new applications in fields where open source labelled datasets are scarce. Space systems engineering is a field with limited access to large labelled corpora and a need for enhanced knowledge reuse of accumulated design data. Transformers models such as the Bidirectional Encoder Representations from Transformers (BERT) and the Robustly Optimised BERT Pretraining Approach (RoBERTa) are however trained on general corpora. To answer the need for domain-specific contextualised word embedding in the space field, we propose SpaceTransformers, a novel family of three models, SpaceBERT, SpaceRoBERTa and SpaceSciBERT, respectively further pre-trained from BERT, RoBERTa and SciBERT on our domain-specific corpus. We collect and label a new dataset of space systems concepts based on space standards. We fine-tune and compare our domain-specific models to their general counterparts on a domain-specific Concept Recognition (CR) task. Our study rightly demonstrates that the models further pre-trained on a space corpus outperform their respective baseline models in the Concept Recognition task, with SpaceRoBERTa achieving significant higher ranking overall.

Sentiment Analysis Using Multi-Head Attention Capsules With Multi-Channel CNN and Bidirectional GRU

Existing text sentiment analysis methods mostly rely on a large number of language knowledge and sentiment resources. This paper proposes the Multi-channel convolution and bidirectional GRU multi-head attention capsule (AT-MC-BiGRU-Capsule), which uses vector neurons to replace scalar neurons to model text emotions, and uses capsules to characterize text emotions. In addition, traditional methods cannot extract the multi-level features of text sequence well. Multi-head attention can encode the dependencies between words, capture sentiment words in text, and using Convolutional Neural Network (CNN) and Bidirectional gated recurrent unit network (Bi-GRU) to extract local features and global semantic features of text respectively, the global average pooling layer is introduced to obtain the multi-level feature representation of the text sequence more comprehensively. This paper selects three English datasets and one Chinese dataset in the general corpus of sentiment classification to conduct experiments, and achieves better results than other baseline models.

Синтетические и аналитические аспектуальные формы монгольского глагола в количественном освещении

Goals. The article attempts to analyze the synthetic and analytical aspectual forms of the Mongolian verb in a quantitative perspective. Materials. The calculation was done on the materials of the General Corpus of the Mongolian Language, and the degree of use of aspectual forms is studied. Results. The paper discusses the issue of distinguishing between synthetic and analytical forms is discussed, delineates certain classes of forms. The following subsystems are distinguished in the system of synthetic forms: 1) subsystem of synthetic verb forms of participial representation, 2) subsystem of synthetic verb forms of predicative-indicative representation, 3) subsystem of synthetic verb forms of adverbial (converbial) representation, 4) subsystem of synthetic verb forms of stem-formation. Within the framework of the generalized system of verbal grammatical forms, analytical aspectual forms are also distinguished. The work characterizes some most common analytical aspectual forms. The most common of them is the analytical form of the ‘progressive’ (// ‘continuative’) formed by combining the synthetic taxis form of the congressional participle (with - ж) (CVB.CNGR) with the connective (existential) verb бай-(х) (lit. ‘ be’). A fragment of the table of the frequency of grammatical forms included in the (analytically expressed) grammeme ‘progressive’ is shown.

Структурно-вероятностная модель монгольской грамматики и измерение употребительности обобщенных грамматических единиц

Syntactic molecule (SM) is the minimum functional and syntactically autonomous unit able to serve as an answer to the question. Goals. The paper formulates the principles of identification of the SM. Results. The main features of grammatical tagging in the General Corpus of the Mongolian Language GCML-3b (compared to earlier versions GCML) are delineated: introducing of the basic vocabulary, principles of monosemization, principles of categorial-semantic notation, main distinctive features of grammar notation (part-of-speech splitting of the indeclinable words and splitting of the nominative). The principles of ordering of morphological structures in a quantitative model of the Mongolian language, as well as the interpretation of suprasegmental markers in written text are discussed. In conclusion, a brief survey of the general structure of the quantitative model of the positional-morphological grammar of the Mongolian language is discussed.

The scope of criminal liability for misappropriation of authorship in EU countries: comparative analysis

Intellectual property legal protection is undoubtedly one of the most important factors and conditions of effective economic, social and cultural development in modern society. According to researchers, absolute majority of countries in the world have set criminal liability for certain crimes against intellectual property rights, including all of the European Union (hereinafter – EU) countries. One of those crimes is misappropriation of authorship. Yet the criminal laws of EU countries criminalise misappropriation of authorship very differently - some protect not only author rights but also related rights, the conditions for criminal liability in the general corpus delicti are of a very different scope as well as the punishments for those crimes differ significantly. This analysis will present the scope of criminal liability in all the EU countries, including the reasons why, as well as will try to find the answer whether ways of coping with these difficulties exist.

Doing Corpus Linguistics: Toward a Conceptual Framework for Indicator of Gender in English Language and Education

Technology plays a pivotal role in the ESL teaching and education sector. In language teaching, gender and language research mostly favors the idea of potential differences in language use between men and women. This paper explores different indicators of gender in the writing of males and females in a large subset of the British National Corpus (BNC) covering the domain of fiction with the application of the Corpus tool. Robin Lakoff's four key linguistic terms that mark female language have been used as benchmarks against which the study has been conducted. Previous researchers like Argamon, Koppel, and Shimoni claim that females use more pronouns and a smaller number of nouns as compared to men. The hits and frequencies of Lakoff's terms and researchers' claims have been checked on BNC to get at the empirical findings. Taking general corpus BNC, corpus research method has been used to answer the research questions. The study found a substantial difference in the documents authored by male and female written text. It was also found that females use many more pronouns and males use many more nouns. Assumptions made regarding Lakoff's terms have been partially substantiated since the results vary a little concerning the use of empty adjectives like 'cute' and 'divine'. The work is a valuable addition to the existing corpus of knowledge about gender differences in language and it provides space for researchers to work in even broader perspectives.

Domain specific word embeddings for natural language processing in radiology

BackgroundThere has been increasing interest in machine learning based natural language processing (NLP) methods in radiology; however, models have often used word embeddings trained on general web corpora due to lack of a radiology-specific corpus. PurposeWe examined the potential of Radiopaedia to serve as a general radiology corpus to produce radiology specific word embeddings that could be used to enhance performance on a NLP task on radiological text. Materials and methodsEmbeddings of dimension 50, 100, 200, and 300 were trained on articles collected from Radiopaedia using a GloVe algorithm and evaluated on analogy completion. A shallow neural network using input from either our trained embeddings or pre-trained Wikipedia 2014 + Gigaword 5 (WG) embeddings was used to label the Radiopaedia articles. Labeling performance was evaluated based on exact match accuracy and Hamming loss. The McNemar’s test with continuity and the Benjamini-Hochberg correction and a 5×2 cross validation paired two-tailed t-test were used to assess statistical significance. ResultsFor accuracy in the analogy task, 50-dimensional (50-D) Radiopaedia embeddings outperformed WG embeddings on tumor origin analogies (p < 0.05) and organ adjectives (p < 0.01) whereas WG embeddings tended to outperform on inflammation location and bone vs. muscle analogies (p < 0.01). The two embeddings had comparable performance on other subcategories. In the labeling task, the Radiopaedia-based model outperformed the WG based model at 50, 100, 200, and 300-D for exact match accuracy (p < 0.001, p < 0.001, p < 0.01, and p < 0.05, respectively) and Hamming loss (p < 0.001, p < 0.001, p < 0.01, and p < 0.05, respectively). ConclusionWe have developed a set of word embeddings from Radiopaedia and shown that they can preserve relevant medical semantics and augment performance on a radiology NLP task. Our results suggest that the cultivation of a radiology-specific corpus can benefit radiology NLP models in the future.

“Maleducados/Ill-mannered”during the #A28 political campaign on Twitter

AbstractThis paper approaches the study of conflict through an examination of Spanish metapragmatic labels and comments of impoliteness on Twitter. The aim is twofold. It first aims to confirm the attributed importance of the labelmaleducado/ill-mannered in the specific context of Twitter and of digital discourse more generally, on quantitative and comparative grounds; then, it investigates this label, and the metapragmatic comments where it occurred, in a contextualized corpus of tweets compiled during the political campaign of Spain’s General Elections of April 28, 2019. The study draws from fivead hoccorpora specifically compiled from Twitter, and a general corpus of Spanish digital discourse provided by Sketch Engine. The analysis adopts a corpus-based metapragmatic approach, which combines quantitative and qualitative methods. Findings revealed thatmaleducadowas the most frequent metapragmatic label under scrutiny in the Twitter corpora and motivated the subsequent study of lay conceptualizations of this term.

Schadenfreude: Malicious Joy in Social Media Interactions.

The paper presents a model of Schadenfreude, pleasure at another’s misfortune, resulting in a typology of cases of this emotion. Four types are singled out: Compensation, Identification, Aversion, and Injustice Schadenfreude. The typology is first tested on a corpus of 472 comments drawn from three social media, Facebook, Twitter and Instagram. Then a specific corpus of comments is collected and analyzed concerning a specific case of Injustice Schadenfreude, the posts concerning Brexit, United Kingdom leaving the European Union. From the analysis, it emerges that spatial or factual closeness does not look necessary to feel Schadenfreude. Finally, a lexicometric automatic analysis is conducted on the general corpus of Italian comments collected using several hashtags and enriched by comments about the fire of Notre Dame, showing how even complex emotions like Schadenfreude can be automatically extracted from social media.

Evolutionary Optimization of Ensemble Learning to Determine Sentiment Polarity in an Unbalanced Multiclass Corpus.

Sentiment polarity classification in social media is a very important task, as it enables gathering trends on particular subjects given a set of opinions. Currently, a great advance has been made by using deep learning techniques, such as word embeddings, recurrent neural networks, and encoders, such as BERT. Unfortunately, these techniques require large amounts of data, which, in some cases, is not available. In order to model this situation, challenges, such as the Spanish TASS organized by the Spanish Society for Natural Language Processing (SEPLN), have been proposed, which pose particular difficulties: First, an unwieldy balance in the training and the test set, being this latter more than eight times the size of the training set. Another difficulty is the marked unbalance in the distribution of classes, which is also different between both sets. Finally, there are four different labels, which create the need to adapt current classifications methods for multiclass handling. Traditional machine learning methods, such as Naïve Bayes, Logistic Regression, and Support Vector Machines, achieve modest performance in these conditions, but used as an ensemble it is possible to attain competitive execution. Several strategies to build classifier ensembles have been proposed; this paper proposes estimating an optimal weighting scheme using a Differential Evolution algorithm focused on dealing with particular issues that multiclass classification and unbalanced corpora pose. The ensemble with the proposed optimized weighting scheme is able to improve the classification results on the full test set of the TASS challenge (General corpus), achieving state of the art performance when compared with other works on this task, which make no use of NLP techniques.

Caracterização das teses e dissertações brasileiras na área de Enfermagem em Unidade de Terapia Intensiva

Objetivou-se caracterizar a produção de teses e dissertações brasileiras na área de enfermagem em Unidade de Terapia Intensiva. Estudo de revisão bibliométrica realizado no mês de outubro de 2018 na Plataforma Sucupira. Utilizou-se o descritor “Unidades de Terapia Intensiva” e incluíram-se teses e dissertações publicadas nos anos de 2013 a 2017. Os dados foram tabulados no Microsoft Excel® e os resumos processados no Interface de R pour les Analyses Multidimensionnelles de Textes et de Questionnaire (IRAMUTEQ). A população constou de 181 produções, das quais 71,8% eram dissertações, 20,6% teses e 7,6% monografias de mestrados profissionais. As regiões do Brasil com maiores percentuais de produções foram a Sudeste (44,0%) e Sul (23,2%). As principais áreas de concentrações em UTI foram: “Enfermagem no contexto social brasileiro” e “Saúde e Enfermagem”. No IRAMUTEQ o corpus geral foi constituído por 181 dissertações e teses, 1714 segmentos de textos e 60.838 ocorrências. Aproveitaram-se 1584 (92,42%) segmentos, configurando-se em corpus significativo. A Nuvem de Palavras destacou a relação entre a terapia intensiva e a “equipe de enfermagem”, a “segurança” e a “segurança do paciente” os diferentes grupos de pacientes desta assistência: “neonatal”, “criança” e “adulto”. A terapia intensiva é um cenário rico e complexo para a realização de pesquisas e os resultados das produções contribuem para a melhoria da segurança do paciente e da assistência de enfermagem.

ChoCo: a multimodal corpus of the Choctaw language

This article presents a general use corpus for Choctaw, an American indigenous language (ISO 639-2: cho, endonym: Chahta). The corpus contains audio, video, and text resources, with many texts also translated in English. The Oklahoma Choctaw and the Mississippi Choctaw variants of the language are represented in the corpus. The data set provides documentation support for this threatened language, and allows researchers and language teachers access to a diverse collection of resources.

Sentiment Analysis using Multiple Word Embedding for Words

Nowadays users express their opinions on different websites like e-commerce and special review websites. Analyzing customers' opinions and their responses is important for decision making. So the researchers worked on analyzing these reviews automatically using a classical machine learning approach like Support Vector Machine (SVM) and various modern deep neural networks. For these networks, words are represented by using vectors called word embeddings. The required word embeddings are taken from pre-trained Word2Vec or learned from a corpus of the given main task. But, each method has its demerits. In the case of pre-trained word embeddings, embeddings are learned from large general corpus so these embeddings are not task specific. While in the case of learning words from the corpus of the main task, it does not reflect the true semantics. To deal with these problems, we have proposed an embedding developer model. This model develops task specific word embedding which also reflects true semantics. Task specific word embeddings are generated from the given corpus using the embedding layer in Keras. It builds the embeddings by considering relationships between words in the window. While true semantics are taken from Word2Vec embeddings. The proposed model combines these two embeddings to generate true semantics and task specific word embeddings. Result analysis shows that the proposed system works better on many benchmark datasets.

Using word embeddings for library and information science research

In the research field of library and information science (LIS), pattern recognition and machine learning approaches have often been adopted to analyze massive amounts of textual data in digital libraries. In recent natural language processing, word embeddings have attracted much attention because they can capture semantic meanings and contexts. This paper presents a short survey of how word embeddings have been used in LIS research, especially focusing on articles published in LIS journals between 2009 and 2019. Our simple bibliographic analysis showed that at least 15 LIS journal papers used word embeddings. These 15 papers were briefly described under the following categories: knowledge extraction and visualization from scholarly data, classification in scholarly data, and others that focus on more general corpora.

An acoustic phonetic description of Nungon vowels.

This study is a comprehensive acoustic description and analysis of the six vowels /i e a u o ɔ/ in the Towet dialect of the Papuan language Nungon ⟨yuw⟩ of northeastern Papua New Guinea. Vowel tokens were extracted from a corpus of audio speech recordings created for general language documentation and grammatical description. To assess the phonetic correlates of a claimed phonological vowel length distinction, vowel duration was measured. Multi-point acoustic analyses enabled investigation of mean vowel F1, F2, and F3; vowel trajectories, and coarticulation effects. The three Nungon back vowels were of particular interest, as they contribute to an asymmetrical, back vowel-heavy array, and /o/ had previously been described as having an especially low F2. The authors found that duration of phonologically long and short vowels differed significantly. Mean vowel formant measurements confirmed that the six phonological vowels form six distinct acoustic groupings; trajectories show slightly more formant movement in some vowels than was previously known. Adjacent nasal consonants exerted significant effects on vowel formant measurements. The authors show that an uncontrolled, general documentation corpus for an under-described language can be mined for acoustic analysis, but coarticulation effects should be taken into account.