Publications & research

Published and presented at DH2025, this work examines textual transformation in Lady Gregory's retellings of Irish legend through computational comparison. The study uses quantitative text analysis to investigate changes in language, style, and textual structure while connecting computational measurements with questions of literary adaptation and transmission.

Presents a machine-learning framework for identifying sentiment and the reasons associated with those opinions in social-media discourse about the US healthcare industry. The approach combines NLP, sentiment analysis, clustering, and topic modelling to extract interpretable patterns from public discussion.

Presented an evaluation of rule-based and large-language-model-assisted approaches to extracting structured lexical information from Hiberno-English dictionary material. The work considers how extraction quality can be assessed across structurally different dictionary entries.

Published and presented at CASCADE × MECANO Convention 1, this work proposes a multidimensional framework for describing relationships between parallel texts by separating questions of authorship from the degree and nature of textual transformation. The model connects literary-theoretical accounts of textual relation with computational approaches to similarity, reuse, adaptation, and transformation.

Investigates how meaningful relationships between texts can be retrieved and evaluated when lexical overlap progressively disappears through paraphrase, obfuscation, adaptation, and allusion. The research combines large-scale semantic retrieval, relation judgement, benchmark construction, and evaluation across a spectrum of textual transformation.

Develops interactive methods for exploring large candidate spaces of textual relations and moving between aggregate patterns and close-reading views. Current work examines retrieval at sentence, paragraph, and chapter level.

Research conducted during the Nimbus secondment examined failures that arise when conversational AI systems translate human instructions into physical action. The work focused on maintaining grounded, action-critical information across planning and tool-use stages.