67 and 85 lexical bundle (LB) types, as well as 2,595 and 3,072 tokens, were found in the Lower-Level Corpus (LLC) and Higher-Level Corpus (HLC). HLC writers were more frequent users of all LB lengths (types and tokens). When evaluated in terms of token counts, these differences were statistically significant. LLC writers used more LBs that overlapped with the provided prompt, indicating a greater dependence on the prompt for language used in their essays. Functional classifications indicated that HLC writers made statistically greater use of all major functional categories. Furthermore, LLC writers had a greater percentage of discourse-organizing structures, with HLC writers holding a slightly higher percentage of stance bundles. These results have been submitted for publication and are also being shared with teachers to help students better understand which structures to avoid/make greater use of in their academic English writing.

This study used academic essays from the Academic Writing and Discussion in English (AWADE) program to create a graded corpus that can now be used for research purposes and program development. This corpus consists of over 200 essays, and each essay was graded by 3 trained raters with academic English teaching experience.All essays from students who had previously submitted a research ethics approval form were digitally downloaded. Next, each digital file was reviewed to ensure no identifying materials (e.g., student names, ID numbers) were included. After removing any writing that could identify particular students, each essay was re-examined to eliminate erroneous characters (e.g., @, *, ~) that are sometimes included because of the downloading method used to gather essays. Finally, overly long/short essays were eliminated to ensure a relatively narrow range of essay lengths. This was important since similar average essay lengths can lead to improved analysis in later stages and more insightful comparisons between writer groups. After preparing the essays for analysis, each essay was evaluated by a minimum of three trained raters. Ratings took place over an approximately 1-month period.

This study used a combination of manual and automated analyses to examine how writers from various first language (L1) backgrounds make use of metadiscourse in their second language (L2) English academic writing. The Authorial Voice Analyzer (AVA; Yoon, 2017), a natural language processing tool for the analysis of interactional metadiscourse, was the main means of data extraction. Initial analyses based on a self-assembled corpus of L1 Arabic, Chinese, and French writing produced in an English as a Second Language context yielded no major differences in metadiscourse. However, this may have been influenced by the frequent presence of in-text citations used by these writers. Therefore, writing from an alternative corpus of L2 English writing, the International Corpus Network of Asian Learners of English (ICNALE) was also analyzed. This corpus produced several noteworthy results, including significant L1-related differences in the use of metadiscourse across all major functional categories (hedges and boosters, attitude markers, self-mentions and reader pronouns) by L1 Chinese, Japanese, and Korean writers.  

The present study used manual analysis of essays from 8 first language (L1) groups sourced from the International Corpus Network of Asian Learners of English to uncover statistically significant differences in the use of linking adverbials (e.g., furthermore, on the other hand, in addition) by each L1 group. The first manuscript from this data set (Appel, in press) targeted L1 Chinese, L1 Indonesian, and L1 Korean writers of second language (L2) English with major findings including relative overuse of resultative linking adverbials (e.g., therefore) by L1 Korean writers, transitions (e.g., nowadays) by L1 Chinese writers, and underuse of contrastive linking adverbials (e.g., however) by L1 Indonesian writers of L2 English. Since the data extraction went ahead of schedule, I was able to collect sufficient data to begin work on a second manuscript targeting L1 Chinese, L1 Arabic, and L1 English writers. This study is currently in preparation and I am to submit this second manuscript by the end of the summer. I believe I should also be able to start working on a third manuscript from this same data set over the summer break. 

The present study used a collection of over400 automated single and multi-word measures to explore their relationship withdifferences in L2 English writing proficiency based on a graded corpus of 496writing samples from a widely used language proficiency test. After running correlationsbetween the full range of measures and the score awarded to each essay, we seta conservative minimum correlation cut-off of .30. To account formulticollinearity between significantly correlated predictor variables, in anycase where two predictor variables held a minimum correlation of .70 with eachother, only one variable was retained. Finally, multiple regression was used tobetter understand which variables could account for differences in writingproficiency. The most parsimonious regression yielded a model with fourpredictor variables, total word count,orthographic neighborhood frequency, lexical decision time, and word naming response time accounting for36% of total variance in writing proficiency scores. These results provide insight into the value of various linguistic features in assessments of L2 English writing proficiency.