Meet MATT: a human-auditable tool for sentence-sampled analysis of English article use

Abstract

English articles (a/an, the, zero article) are central to second-language (L2) accuracy research but remain difficult to analyze because appropriate use depends on semantic, syntactic, and discourse-level interpretation. This paper introduces MATT (Model for Article Tracking & Typology), a web-based tool designed to produce sentence-sampled, human-auditable annotations of article use in learner writing. MATT accepts pasted text or uploaded files, segments the text into sentences, and evaluates one article-sensitive noun phrase per sentence. For each evaluated sentence, it generates a structured record that includes the target noun phrase, observed and expected article types, a broad error class aligned with omission, overuse, and misuse, a suggested minimal correction, a heuristic review-priority score, and a short rationale. Because MATT does not exhaustively annotate all noun phrases, its outputs should be interpreted as a sampled annotation layer rather than a complete corpus annotation. Accordingly, any accuracy values derived from MATT output, including MATT-derived Suppliance in Obligatory Contexts (MATT-SOC) and MATT-derived Target-Like Use (MATT-TLU), should be understood as sampled indicators based on the evaluated noun phrases. The interface provides correction views, summary tables, visualizations, and downloadable CSV files so that researchers can inspect individual decisions, recompute summary values, and document the scope of analysis transparently. The paper explains the tool workflow, interpretation of MATT-derived outputs, and recommended reporting practices for exploratory article-use analysis.