American Association for Applied Linguistics
2022

In 2022 I co-presented, with Dr. Elena Cotos, a paper on the AFLEX project: An interdisciplinary approach to developing an automated functional language extraction (AFLEX) system to transform the translation of STEM research to society at the American Association for Applied Linguistics conference. 

Abstract: Innovative technologies for research writing rely on automated text analysis models trained to identify linguistic and rhetorical patterns across disciplines (e.g., Fiacco et al. 2019). Similar models are needed for other applied natural language processing applications, such as to facilitate systematic reviews. The systematic review synthesizes research on a specific topic from academic publications. A major challenge for automating research synthesis is identifying study design details, which are essential to assessing the validity of research. Current state-of- the-art text mining tools lack this level of granularity because they rely on keyword detection combined with statistical models and machine learning algorithms.

This interdisciplinary project aimed to determine whether functional-semantic patterns could feasibly replace keyword detection in a text analysis/classification model for automatically identifying study design details. In collaboration with domain experts in biomedical sciences, a discipline-specific corpus of 100 published research articles was compiled. The experts manually extracted 2085 passages from these texts and categorized them as 8 study design elements (e.g., allocation) and 25 design options (e.g., random, systematic, haphazard). A random sample of passages were analyzed inductively to identify and define linguistic patterns and their lexico-grammatical instantiations. Eight functional-semantic patterns emerged: alternative, comparison, coordinate structure, duration, manner, quantity, substance, and temporal. Then 738 passages categorized as design elements and options were coded in terms of functional-semantic patterns, which yielded 6033 linguistically coded units. Results of statistical tests of independence indicate a relationship between study design categories and functional-semantic patterns. This study is unique for establishing the use of functional-semantic patterns in place of keywords as the basis for a text mining method. Future research will involve collaborating with experts in machine learning to train and evaluate a text analysis/classification model for automatic extraction of research design information based on the functional-semantic patterns.