Applied Corpus Linguistics最新文献

Previous research examined L2 interaction by describing salient features exhibited in different patterns of peer interaction. These studies mostly used qualitative methods and focused on the collaborative aspect of such construct (Galaczi, 2008). The present study adopts a quantitative approach to explore and describe L2 interaction, utilizing the data of the Corpus of Collaborative Oral Tasks (CCOT). Specifically, it measures pairs’ interaction by creating a composite score of interactivity to understand the relationship between the dyads' degree of interactivity and their use of lexico-grammatical features as well as their L2 fluency. Pearson's correlation tests showed weak to moderate positive relationships between interactivity and discourse particles, response forms, wh-questions, and second person pronouns. Additionally, the tests revealed weak negative relationships between interactivity and both nominal forms and hesitations. Furthermore, revealing moderate relationships, Pearson's correlation tests showed that interactivity was associated with more fluent L2 speech, where learners of higher interactivity levels tended to produce fewer silent pauses and faster speech rates. The study provides insights for scholars interested in L2 interaction. It suggests that some linguistic features were not only associated with collaborative behaviors (as reported in the literature) but also with interactivity as broad aspect. Furthermore, it provides a description of how the act of turn taking might potentially serve the fluency of higher interactivity students, warranting further investigation of turn frequency among L2 test takers as test raters might potentially be influenced by the test candidates’ fluency. Finally, it reports that L2 interactivity exhibited a relationship pattern with linguistic features that resembles patterns reported in the literature of studies on native speakers of English.

Register is among the most important predictors of linguistic variation. In a register such as instructor feedback, linguistic features have particularly high stakes, as they can make feedback more clear, detailed, and/or (de)motivating. Mitigation strategies (i.e., the use of hedges and other softeners) are frequently found in instructor feedback and are particularly influential in terms of the feedback's effectiveness. This study compares the patterns of mitigation strategies used in written and spoken feedback to gain insights into register variation. Written comments (provided electronically) and spoken comments (provided through screencast feedback, in which instructors share verbal feedback along with a screenshare of the student's essay) in the Writing Feedback Corpus (WFC) were analyzed. 1,568 comments across these registers were manually coded for mitigation within head acts (core speech acts) and external modification in the surrounding discourse. Strategies were compared quantitatively using key feature analysis (Egbert & Biber, 2023). The findings indicate that feedback registers promote the use of different mitigation strategies and external modification strategies, with written feedback favoring interrogative syntax and unmitigated forms and spoken feedback favoring personal attribution, hedges, and the nursery we as well as the external modifiers minimizer, positive comment, and reason. Implications for providing feedback on student writing are highlighted.

Corpus findings are only useful if the corpus adequately represents the content and language of the target domain; yet few studies evaluate or report representativeness. This paper argues that corpus linguists should focus explicitly on the validation process. It introduces the innovative concept of a Representativeness Argument, which is an explicit statement of reliability and validity to enable defensible applications of a corpus for a specifically defined purpose and audience. Adapted from Toulmin's (1958/2003) argument model, its originality lies in its attention to both target domain and linguistic representativeness, and in the critical role played by expert judgements. To illustrate this approach, I present a representativeness argument for the 1.98-million-word ‘DSVC-IL’ corpus, which was compiled to investigate the discipline-specific vocabulary required for reading postgraduate International Law texts. The corpus is demonstrated to adequately represent target domain content, established by analysing modules and reading lists, and confirmed by experts. The language is shown to adequately reflect the domain through analysis of a 1026-flemma Single Word List, extracted using measures of frequency, keyness, range and evenness of distribution. List items are evenly-distributed in randomly-split corpus halves (r_s=.98, p<.00). The list provides similar coverage of the DSVC-IL (26.37%) and other texts from the domain (23.87%). Moreover, Law experts confirmed the majority of list items were Law words. Together, the evidence supports the usefulness of the corpus and list for its explicitly defined purpose.

With the proliferation of large corpora and the availability of sophisticated corpus-analysis tools, the number of corpus-based word lists targeting different types of vocabulary has rapidly increased during the last 20 years. This wide variety of lists has caused problems for practitioners, for whom it is not always easy to decide which list is most useful for their purpose and context. Given the paucity of systematic guidance on how to evaluate word lists, this study aimed to construct an evaluation tool that is based on Nation's (2016) framework of critiquing word lists, but is reformulated for a different purpose and for different target users, in order to increase the applicability of information derived from corpus analysis (the word lists). Constructed based on a thorough literature review, and informed by practitioners’ views and uses of word lists, along with consultations with ELT practitioners and word list experts, the tool targets ELT practitioners such as teachers, curriculum and assessment coordinators, and materials developers involved in directing vocabulary acquisition. The tool caters to practitioners with different levels of expertise and knowledge—especially those who are unfamiliar with the intricacies of developing corpus-based word lists. This paper documents the development of the initial version of the evaluation tool, as well as its first iteration, drawing upon the insights of both word list experts and practitioners in ELT.

This study investigated the constructional knowledge development of L1 Indonesian by examining nineteen Verb-Argument Constructions (VACs). The VACs examined in the present study are a verb pattern, followed by a preposition and a noun phrase, for example, “V about N” as in “He talked about the progress”. This study used the Indonesian subset of the Education First Cambridge Open Language Database (EFCAMDAT) corpus from beginner to advanced levels (CEFR A1 to C1; Council of Europe, 2001). This dataset comprises 2943 writing texts (224,763 words) from 623 learners. Frequency analysis of types and tokens was conducted to examine the distribution of the 19 VACs in learner writings across levels. Growth analyses were conducted to investigate the verbs that learners most frequently associated with the most productive VACs. Correlational analyses were conducted to explore how closely related the verb-VAC associations between proficiency levels and the verb occupants in the associations. The results indicate that learners’ constructional knowledge development was implied by: (1) the frequency increase in types and tokens of VACs from lower to higher proficiency levels, (2) the variety of verbs associated with VACs, and (3) the construction schematicity increase indicated by the use of general to more specific verb productions distinct to proficiency levels. The results suggest that English language learners need more exposure to lexicogrammatical features to facilitate VACs acquisition and usage.

This paper presents the third stage of a larger research project examining perceptions of the International Baccalaureate (IB) to better understand its growing influence on education systems around the world. The first two stages involved a synchronic and diachronic analysis of IB discourse in a 27 million word specialized corpus of global press articles, created as an unsolicited window into public opinion (Mautner, 2008). The present study uses the same corpus to explore how the IB is represented in different countries, what values and attitudes may be associated with it, and how it interacts with other global education actors. Bottom up and top down methods from corpus-assisted discourse studies (CADS) were used to analyze 34,104 newspaper articles from 56 countries. Frequency, collocation and concordance analyses revealed four dominant discourses of deficiency connected to national education systems in countries across the ideological spectrum that helped to legitimize the inclusion of private actors in the provision of education. Results also showed unique discourses associated with the IB in North America, thereby highlighting the key role that this region plays in the IB world.

This paper reports the process of collecting a corpus with examples of anti-vaccine discourse and the results of its linguistic analysis. The overall aim of the project is to help public health authorities to improve their communication campaigns by better understanding the conditions for misinformation spreading via social media. More specifically, this paper analyses linguistic properties of a corpus of prominent misinformation channels in Telegram as compared against a more general COVID corpus as well as against a general purpose English corpus. For this paper, the quantitative analysis relies on corpus querying to identify the most recurrent discourse patterns related to COVID vaccines. We use the appraisal framework to analyse the patterns with respect to the attitudes conveyed in the messages. We have also applied an automatic AI classifier to predict communicative functions of these texts. This allows us to examine them more closely through the use of simple lexicogrammatical features following Biber, as well as their ideational processes following Halliday. The findings show that common collocations in the Telegram corpus containing misinformation draw on three attitudes: fear, insecurity, and mistrust in COVID vaccines which are discursively constructed to promote vaccine hesitancy among social media users. Furthermore, the misinformation messages tend to occur more often in such communicative functions as promotional texts, news reporting, and text expressed as presenting reference information.

This work tracks public health messaging and evidence of stability and change in corpora of the Centers for Disease Control and Prevention (CDC), World Health Organization (WHO), and National Health Service (NHS) official account tweets throughout 2020. Using corpus-based methods, including keyword analysis, major similarities and differences are identified across tweets by each organization over time. Larger macro-level and micro-level discourses and linguistic patterns are revealed, with specific applications relevant to public health and governmental messaging, especially regarding risk and health communication. Findings include the NHS providing the most comprehensive and varied messaging out of each organization, including references to recommended actions, communities and individuals, and information. The WHO focuses predominantly on cases and region-specific information, while the CDC includes a variety of information, with a US-internal focus. Applications include further recommendations for public health communication, including the necessity of diverse linguistic patterns and interactive messaging tactics for governmental organizations.