The following are current on-going corpus-related research projects and publications by the directors and students of the Second Language Studies Corpus Lab, along with additional collaborators in some cases. Click on any of the titles below to read a description of the project.

Syntactic complexity across modalities in L2 learners’ production data (Eunmi Kim)

Previous studies using the Register-Functional (RF) approach (Biber et al., 2021) have shown that complexity features vary across register (Gray et al., 2019), disciplines (Staples et al., 2023), tasks (Kim et al., 2025), and genres (Yoon & Polio, 2017). However, most of them have explored complexity features in written productions, rarely comparing them across modalities. To date, no empirical research has investigated the development of grammatical complexity across proficiency levels in both spoken and written modalities. Therefore, I investigate how Korean L2 learners develop complexity features across L2 proficiency levels in both spoken and written data. I used speeches (520 samples) and essays (600 samples) produced by Korean L2 learners, ranging from beginner to upper-intermediate levels (A2, B1 low, B1 high, and B2), from the International Corpus Network of Asian Learners of English. Syntactic complexity was measured using lexico-grammatical features at three levels: finite dependent clauses, non-finite dependent clauses, and dependent phrases. The developmental complexity tagger (Gray et al., 2019) was used to identify these features associated with each stage of the developmental framework.  

The preliminary findings suggest that L2 learners at the A2 and lower B1 levels used more conversational complexity features in both speech and writing, whereas more advanced L2 learners at the B2 level reduced the use of finite dependent clauses and increased the use of noun phrases in writing. L2 learners at the upper B1 level started to show a gradual decrease in finite dependent clauses and an increase in non-finite dependent clauses. A gradual transition from clausal to phrasal would be demonstrated in the writing produced by L2 learners at the A2 to B2 levels. This study is expected to contribute to the theoretical development of more accurate interpretations of complexity features across modalities and proficiency levels. The findings can provide more modality-specific proficiency scales, rather than treating complexity features as a single scale for L2 development. 

In academic writing, formulaic language plays an important role, signaling writers’ membership within disciplinary discourse communities. Among such formulaic sequences, phrase-frames (p-frames)—discontinuous multiword sequences with an open slot (e.g., the [aim, purpose, goal] of this study)—have been increasingly recognized as an important phraseological feature of academic discourse and a pedagogically useful resource (e.g., Lu et al., 2018). Recent corpus-based studies have revealed notable linguistic and structural differences between quantitative and qualitative studies, which might stem from the two paradigms’ distinct epistemological stances and methodological practices (e.g., Candarli & Jones, 2019; Gray, 2013). Despite these documented differences, how p-frame use varies across the two paradigms remains underexplored.

This study examines how the use of p-frames differs across research paradigms in a subfield of applied linguistics (i.e., language learning and teaching). Two specialized corpora comprising quantitative and qualitative research articles (2022-2024) were compiled from four leading journals publishing research on language learning and teaching (Applied Linguistics, Journal of Second Language Writing, The Modern Language Journal, and TESOL Quarterly). Five- and six-word p-frames were extracted using kfNgram (Fletcher, 2011), which identified p-frames directly from the complete sets of five- and six-grams. These candidate p-frames were subsequently filtered manually through concordance analyses in AntConc (Anthony, 2024) to exclude frames that were semantically non-coherent. The resulting p-frames were analyzed for internal variability and predictability and then classified according to structure and discourse functions.

The preliminary analyses showed that quantitative studies exhibit far greater formulaicity than qualitative studies. (The exact structure and functions of the p-frames are still being analyzed.) The findings of this study add to the growing body of research on disciplinary and paradigm differences in academic publications. This study also has pedagogical implications for English for Academic Purposes, especially for early-career multilingual scholars seeking to develop knowledge of discipline-specific linguistic conventions.

One strand of corpus-based English for Academic Purposes (EAP) research involves cross-disciplinary comparisons to reveal how disciplinary conventions shape linguistic and rhetorical practices of academic genres with the goal of applying such research to teaching EAP. Yet how researchers choose to divide the landscape of academic writing has not been deeply investigated, and such methodological choices have implications for teaching, research, and theory building. 

We collected corpus-based EAP studies published between 2014 and December 2025 from five leading EAP and applied linguistics journals (because a search of databases resulted in excessive irrelevant studies). The final corpus consisted of nearly 100 articles, including a notable recent increase in 2025. We coded researchers’ self-reported justifications introducing a typology consisting of three overarching categories: practical justification (based on convenience or resource availability); empirical justification (feature-focused; discipline-focused; both-focused); and theoretical justification (discipline-focused; discipline and feature-focused).  

Our findings are that practical justification dominates, typically driven by convenience or available resources, followed by empirical, and justifications. Many studies that used prior empirical studies to justify examining specific features but consistently lacked explicit justification for discipline selection. In addition, theoretical justification studies often rely uncritically on the hard/soft disciplinary distinction (e.g., Becher and Trowler, 2001) without explicitly linking it to the features studied.  

By systematically mapping these patterns and identifying key gaps, this synthesis provides a comprehensive framework and explicit recommendations for strengthening cross-discipline corpus research. Improving methodological rigor, transparency, and theoretical coherence is critical for cumulative knowledge-building in EAP research. 

This project examines how AI-mediated Dynamic Assessment (DA) can support Korean as a Foreign Language learners’ speaking and pronunciation development. It includes three connected components: text-based custom GPT mediation for beginner Korean interaction, pronunciation-focused dynamic assessment of Korean sentence-final intonation, and the development of a classroom-oriented dashboard for Korean speaking and pronunciation practice.

 

Part 1: Custom GPT as Mediator in Korean Dynamic Assessment

Part 1, published in Language Learning & Technology, examined how a custom GPT-based chatbot mediated beginner Korean learners’ development through DA. In a four-week classroom-based study with 10 beginner KFL learners, 33 learner-GPT transcripts were collected, comprising 280 learner-GPT turns and 596 mediation strategies. The findings showed that the chatbot sustained level-appropriate interaction, provided graduated feedback responsive to learner errors and confusion, and supported learner development through conversational, instructional, and developmental mediation. This study provides the foundation for extending AI-mediated DA to additional dimensions of Korean speaking development (Kim).

Part 2: Corpus-Based Analysis of AI-Mediated Dynamic Assessment for Korean Intonation

Part 2 extends AI-mediated DA to Korean pronunciation, focusing on sentence-final intonation contrasts between declaratives and yes/no questions. The study develops an AI-based system that uses automated F0 extraction and a global-local classification approach to analyze learners’ productions of Korean minimal-pair sentences. KFL learners complete pre-test recordings, receive graduated pronunciation mediation through a DA chatbot, and complete post-test recordings with parallel stimuli. Using corpus-based acoustic analysis of learner speech before and after mediation, the study examines patterns of sentence-final pitch movement, boundary-tone realization, feedback uptake, and changes in learners’ production of Korean intonation contrasts.

Part 3: Korean AI Speaking and Pronunciation Dashboard

Part 3 builds on the text-based conversation study in Part 1 and the intonation-focused study in Part 2 to develop a broader dashboard for Korean speaking and pronunciation development. The dashboard will be designed for classroom use by KFL instructors and will allow learners to practice pronunciation features such as sentence-final intonation, segmental contrasts, final consonants, sound changes, fluency, and rhythm, as well as speaking tasks across communicative topics and functions. Learners will receive graduated AI mediation, revise or repeat their productions, and track their progress across speaking and pronunciation targets. With appropriate consent, the resulting audio recordings, transcripts, acoustic measures, AI feedback, learner responses, and revised productions will be organized into an annotated Korean learner speech corpus. Corpus-based analyses of these data will examine pronunciation development, feedback uptake, and changes in learner production over time, while also informing the design of speaking tasks, pronunciation feedback, progress-tracking features, and assessment materials for Korean language instruction.

The quantitative/qualitative divide is frequently assumed rather than empirically examined, particularly in studies of disciplinary writing. Corpus and genre analyses offer insights into epistemological differences across traditions (Gao et al., 2022), but examine only observable textual usage, overlooking the sociocultural aspects of disciplinary practice (Cutting, 2021)—including how scholars themselves understand and negotiate the conventions they follow. This paper interrogates whether the qual/quant divide is reflected in scholars’ rhetorical practices, or whether it operates more as an epistemological identity than as a visible feature of disciplinary writing.

This article reports on a mixed methods study examining scholars’ understanding and usage of two rhetorical moves in applied linguistics research: acknowledging limitations and recommending future research. The current study extends a prior corpus and genre analysis study of these moves in 200 published studies (Montgomery, 2023). That study operationalized a qual/quant binary by excluding mixed methods work, an important decision worth revisiting. Here, I present survey and interview data from 114 applied linguists varying in geographic location, career stage, and methodological orientation. The corpus findings revealed a significant paradigmatic difference: limitations and future directions sections appeared in 94% of quantitative studies but only 76–79% of qualitative studies (p<.001), with quantitative studies also exhibiting greater prominence across four target journals. This motivated the central question of the current study: do scholars’ own understandings of these moves reflect a similar divide?

A thematic analysis (Miles et al., 2020) of written survey responses (n=84) and semi-structured interviews (n=20) produced 27 codes across six themes. Quantitative survey data revealed strong cross-paradigm correlations between scholars’ reported use of the two moves (r=.85, p<.001), suggesting they are deeply intertwined regardless of paradigm. Following a quantitizing approach (Sandelowski et al., 2009), a normalized frequency analysis of qualitatively-derived codes revealed selective but meaningful paradigmatic differences. Qualitative scholars were more likely to view limitations as risky and in need of deeper manuscript integration—reflecting an interpretive orientation toward knowledge as situated and negotiated. Quantitative scholars more often described both moves instrumentally, framing peer review as a compliance exercise. Yet notable paradigm differences appeared in only three of 27 codes. Across both groups, scholars described tensions between authentic critical reflection and performative convention (e.g., “ticking the box” or “covering your ass”). Career stage, editorial experience, and mentoring environment emerged as equally or more influential than paradigm, pointing to a socialization-based explanation that cuts across the qual/quant binary.

The study makes three contributions to this special issue. It triangulates corpus and genre evidence with qualitative scholar perspectives, directly enacting the methodological dialogue the issue seeks to encourage. It finds that excluding mixed methods work in corpus studies may inadvertently reify the binary under investigation, thereby raising broader questions about how research designs construct the divides they claim to describe. Finally, it argues that socialization into rhetorical practice may matter more than paradigm membership, with implications for how graduate programs approach disciplinary writing instruction.

Subordination has been considered one of the criteria for measuring syntactic complexity, and it is often used to assess L2 learners’ proficiency levels (Basterrechea & Weinert, 2017). However, most research on adverbial clauses focused on English L1 production data without considering the role of affective factors. This study fills this gap by examining various factors that affect the order of adverbial subordinate clauses and main clauses in Korean L2 writing from the Gachon learner corpus: learner-related (proficiency levels, autonomy of motivation, self-assessed confidence in writing) and linguistic factors (adverbial clause type and clause length). I used a usage-based approach with the corpus, which recognizes that linguistic structures are shaped by meanings, contexts, uses, and language users, and refrains from dismissing linguistic phenomena as peripheral (Basterrechea & Weinert, 2017).

A generalized linear mixed-effects model (GLMM) with a binomial distribution and a logit link was fitted to examine whether linguistic and learner-related factors predicted adverbial clause ordering. The results showed that the length of the main clause, clause type, learners’ proficiency, and autonomy of motivation significantly affected the ordering of the clause (subordinate-main; SM vs. main-subordinate; MS). Among linguistic factors, longer main clauses significantly increased MS ordering, an L2-like pattern, and L2 learners used this pattern most frequently in causal adverbial clauses. Among learner-related factors, greater autonomy of motivation increased MS ordering, whereas higher linguistic competence reduced it. However, the effects of main clause length were moderated by learners’ autonomy of motivation. This indicates that controlled learners relied heavily on surface structure, whereas autonomously motivated learners relied on discourse processing.

The findings of this study demonstrate how various factors affect adverbial clause ordering and help explain L1 transfer of syntactic structures. This study complements research on individual differences, which is mostly experimental.

The Pennsylvania Dutch Dialects Project (penndutchdialects.org) is the first ever corpus of spoken Pennsylvania Dutch, an under-documented Germanic language spoken by various groups throughout North America. The most well-known of these groups are the Amish and the Mennonites. This project seeks to document the colorful tapestry that is Pennsylvania Dutch by including samples representing as many speaker and language types as possible (e.g., region, affiliation, gender, and age of speakers, etc.). In addition to documentation, we strive to educate any interested parties (the general public, language enthusiasts, teachers of German or Pennsylvania Dutch, the speakers themselves) about where the language comes from, how it has changed over time, and why it is a legitimate language. To do so, we are working to develop outreach materials based on the database of spoken Pennsylvania Dutch that can be used to engage with the language in new ways.  

  • Gagnon, S. (2025). A corpus based exploration of the progressive -ko iss construction in L1, L2, and textbook KoreanDissertation. Michigan State University  
  • Gao, J., Pham, Q, & Polio, C. (2023). The role of theory in structuring literature reviews in qualitative and quantitative research articles. Journal of English for Academic Purposes, 63, 1-13. https://doi.org/10.1016/j.jeap.2023.101243 
  • Gao, J., Pham, Q, & Polio, C. (2022). The role of theory in quantitative and qualitative second language learning research: A corpus-based analysis. Research Methods in Applied Linguistics, 1(2), 1-14. https://doi.org/10.1016/j.rmal.2022.100006 
  • Gao, J., Pham, Q., & Polio, C. (2025). Citations in post-methods sections of quantitative and qualitative research articles in second language learning: A corpus-based study. Journal of English for Academic Purposes74, 1-11. https://doi.org/10.1016/j.jeap.2025.101473
  • Pfau, A. (2024). Construction and Evaluation of Data-Driven Learning Modules for EFL Writers’ Hedging in Academic EnglishDissertation. Michigan State University.
  • Hwang, H.B., & Polio. C. (2023). Text length effects on the reliability of syntactic complexity indices. Research Methods in Applied Linguistics, 2(3), 100085. https://doi.org/10.1016/j.rmal.2023.100085 
  • Montgomery, D. P. (2023). “This study is not without its limitations”: Acknowledging limitations and recommending future research in applied linguistics research articles. Journal of English for Academic Purposes. 65, 101291. https://doi.org/10.1016/j.jeap.2023.101291
  • Pfau, A., Polio, C., & Xu, Y. (2023). Exploring the potential of ChatGPT in assessing L2 writing accuracy for research purposes. Research Methods in Applied Linguistics, 2(3), 100083. https://doi.org/10.1016/j.rmal.2023.100083  
  • Polio, C., & Yoon, H.J. (2020). Exploring multi-word combinations as measures of linguistic accuracy in second language writing. In B. Le Bruyn & M. Paquot (Eds.), Learner corpora and second language acquisition research (pp. 96-120).  Cambridge: Cambridge University Press. https://doi.org/10.1017/9781108674577.006
  • Xu, Yi., Polio, C., & Pfau, A. (2024). Optimizing AI for assessing L2 writing accuracy: An exploration of temperatures and prompts. In C. Chapelle, G. H. Beckett, & J. Ranalli (Eds.), Exploring AI in applied linguistics (pp. 151-174). Iowa State University Digital Press. https://doi.org/10.31274/isudp.2024.154.10