Applied Corpus Linguistics最新文献

The present study explores the development of grammatical complexity in L2 English writing at the beginner, lower intermediate, and upper intermediate levels to see (i) to what extent the developmental stages proposed in Biber et al. (2011) are evident in low-proficiency L2 writing, and if so, what the patterns of progression are, and (ii) whether students gradually move away from speech-like production toward more advanced written production. We use data from COBRA, a corpus of L1 Brazilian Portuguese learner production, along with BR-ICLE and BR-LINDSEI. All the data were tagged using the Biber tagger (Biber, 1988) and the Developmental Complexity tagger (Gray et al., 2019), and subsequently analyzed using a technique developed in Staples et al. (2022) to quantify developmental profiles across levels. The technique considers not only overall change in frequency across levels, but also the incremental variation across each adjacent level (based on % frequency changes). The results show that the features were infrequent overall, with a majority of both clausal and phrasal features exhibiting an increase in frequency across the levels, albeit to varying degrees. This general pattern is contrary to predictions based on findings from previous studies, which found phrasal features increasing in use and clausal features decreasing in use. Nonetheless, for the features associated with each developmental stage, the frequencies generally increased, becoming more similar to advanced written production and more dissimilar to spoken production, as hypothesized in Biber et al. (2011).

Data-driven learning (DDL) refers to the use of corpora by second and foreign language (L2) learners to explore and inductively discover patterns of their target language use from authentic language data without interventions from others. Although previous meta-analyses have demonstrated the positive effects of DDL on L2 learning (Boulton and Cobb, 2017), the number of empirical studies has been increasing since then. Therefore, this study included more recent studies and used meta-analyses to examine the extent to which: (1) DDL exerts an effect on L2 learning; and (2) moderator variables affect DDL's influence on L2 learning. The results demonstrated small to medium effect sizes for experimental/control group comparisons and pre/post and pre/delayed designs. Moreover, the moderator analyses found that moderator variables, such as publication types, learners’ factors, and research designs, influence the magnitude of DDL effectiveness in L2 learning.

This article presents the theoretical basis for corpus linguistics applied to Aeronautical English teaching and assessment followed by practical examples on how to use corpora to develop tasks for both purposes. It originates from the design of two webinars held remotely at the end of 2020, and promoted by the International Civil Aviation English Association. The webinars were targeted at Aeronautical English teachers, material designers, and test developers with little or no previous knowledge of corpus linguistics with the aim of guiding the audience in preparing step–by–step tasks using corpora. We share the work involved in the task design suggested, bridging the gap between research and practice. We conclude by outlining limitations, and suggesting prospects for future research.

This study explores diachronic variation across approximately one hundred years of the newspaper register in US American English from 1920 to 2019 as captured in the Corpus of Historical American English (Davies, 2010). Informed by a similar study of lexical change in British English (Baker, 2011), the analysis identified high-frequency words exhibiting the greatest increases and decreases in use as well as those words demonstrating stability across the four sampling periods: 1920–29, 1950–59, 1980–89, 2010–19. The process to identify words of change and stability began first with the application of a cumulative frequency threshold; coefficient of variance and Kendall's Tau correlation coefficient were then calculated to aid in identification. In other words, the process targeted high-frequency words whose use has demonstrated the greatest change or stability. The discussion presents the three resulting word lists (increasing, decreasing, stable) and reports concordance and collocation analysis of select words from each list to gain insight into the underlying factors informing lexical change and stability.

This paper explores the intersection of data-driven learning (DDL) and generative AI (GenAI), represented by technologies like ChatGPT, in the realm of language learning and teaching. It presents two complementary perspectives on how to integrate these approaches. The first viewpoint advocates for a blended methodology that synergizes DDL and GenAI, capitalizing on their complementary strengths while offsetting their individual limitations. The second introduces the Metacognitive Resource Use (MRU) framework, a novel paradigm that positions DDL within an expansive ecosystem of language resources, which also includes GenAI tools. Anchored in the foundational principles of metacognition, the MRU framework centers on two pivotal dimensions: metacognitive knowledge and metacognitive regulation. The paper proposes pedagogical recommendations designed to enable learners to strategically utilize a wide range of language resources, from corpora to GenAI technologies, guided by their self-awareness, the specifics of the task, and relevant strategies. The paper concludes by highlighting promising avenues for future research, notably the empirical assessment of both the integrated DDL-GenAI approach and the MRU framework.

This study assessed the validity of descriptive findings from corpus linguistics research by analyzing human participants’ performance and perception data. While the stative verb believe usually occurs in the simple aspect, a corpus-based analysis has revealed that believe also occurs in the progressive form in communicative situations conveying a heightened degree of intensity and marked with specific linguistic features such as intensifying adjectives, adverbs of certainty, direct addresses, and others (Gracheva, in press). This study adopted an experimental approach to further assess the link between the progressive form in situations of use conducive to assertive stance and emotional involvement and its surrounding linguistic characteristics. Eighty-six native English speakers were presented with 24 naturally-occurring texts from corpora. Half of the texts involved linguistic features of intensity (progressive aspect condition), while half involved no such features (simple aspect condition). Participants read the texts and selected the form of believe (simple or progressive aspect) which they thought was appropriate in each text. Results showed that participants selected the progressive aspect 47% of the times for the texts featuring language of intensity, while their selection of that aspect was less than 3% in the simple condition texts. Follow-up interviews revealed that participants sensed the intensity conveyed by the texts (e.g., strong emotion, urgency, emphasis), leading to their choice of the progressive over the simple aspect.

This paper describes the design and development of the DAIS-C (Discussing Abstract Ideas in Schizophrenia Corpus), a small, specialised corpus of spoken language in which speakers with a diagnosis of schizophrenia and those with no self-reported psychiatric or neuroleptic history were interviewed on the same topics. The corpus was constructed to allow for comparative analyses of speech behaviour in relation to linguistic creativity and formal thought disorder (FTD), but additional steps were taken to ensure that the corpus could be of use to other researchers and research questions. The present paper covers design decisions relevant to the construction of clinical corpora alongside information about the corpus of potential use to researchers interested in its use.

India has a huge diversity of languages, and Hindi and Marathi are the most spoken languages in the northern and western parts of India. Hindi and Marathi have more than 528 million and 83 million speakers, respectively. The paper describes the development of the Hindi Word Corpus (Hindi WordCorp) and the Marathi Word Corpus (Marathi WordCorp), reporting the frequency of single words (1-gram) used in written texts of the respective languages using the bag-of-words model (BoW). The word frequencies are provided for eleven decades (pre-1920, 1920 to 2020). These texts include books (fiction, non-fiction, history, autobiographies, etc.) and magazines. Academic and reference books were not used. The Hindi WordCorp and Marathi WordCorp used 640 and 712 texts, respectively. An analysis was employed to check whether the texts used were enough to stabilize the rank-order of the total frequencies of the words. Zipf's and Heaps’ law coefficients indicated the sufficiency of the texts. Researchers in various areas like linguistics, social sciences, text mining, machine learning, etc., can use the dataset to answer research questions about language and culture. Some demonstrative examples are provided for using the datasets in the two languages. The dataset is made available on an open data repository. The paper is an account of dataset creation for Hindi and Marathi WordCorp. Hence, no empirical results or conclusions are made based on the data created. A WebApp named Indian Languages Word Corpus (ILWC) has been developed for users. Future directions for text mining and language models are discussed.

YouTube is becoming an increasingly popular entertainment platform, with videos catering to a wide range of interests. If L2 users are to become proficient in the primary form of language, conversation, then the affordances created by YouTube videos containing informal speech could be very useful. In the current study a near-random corpus of 2602 YouTube video transcripts was compiled and 200 randomly selected texts from the Spoken BNC2014 (Love et al., 2017) were used as a reference corpus representing informal spoken English. The texts were tagged with 67 linguistic features as part of an additive multi-dimensional analysis. The dimension scores for each text were used in a cluster analysis to investigate which texts clustered with the Spoken BNC2014 texts. A two-cluster solution was chosen with 666 YouTube texts and 171 Spoken BNC2014 texts in one cluster, and the remaining texts in the other cluster. A small sample of texts from each cluster was analysed in detail. It is shown that this method has the potential to identify videos featuring informal speech and that some videos with similar categories have a very different linguistic style.