Lunchlezing: Automatic CHAT Annotation: A Feasibility Study

During this lecture, Frank Wijnen (Professor of Psycholinguistics, UU) and Mees van Stiphout (Centre for Digital Humanities, UU) will expand on how complete and correct CHAT annotations can be generated on the basis of transcribed actual utterance-correct version pairs.

In an ongoing project, GrETEL 4 has been applied to the grammatical analysis of child language and language produced by people with aphasia, and has been integrated into a clinical tool, SASTA. In order to find accurate parses of non-standard utterances, SASTA requires that transcriptions are annotated in accordance with CHAT guidelines. However, as CHAT annotating is intricate and time-consuming, researchers and clinicians mostly annotate by adding ‘correct versions’ of child utterances.

The present project aimed to generate complete and correct CHAT annotations on the basis of transcribed actual utterance-correct version pairs. The approach has been to align these two representations through weighted edit distance calculation. This has proven successful; the current version of the algorithm succeeds in correctly aligning actual utterances with correct versions, and produces correct CHAT codes. Experiments show that applying SASTA enriched with AuChAnn markedly improves grammatical analysis accuracy. Integrating AuChAnn into SASTA will yield both better results and increased user-friendliness, which is expected to promote implementation of SASTA in clinical practice.

This CLARIAH Lunch lecture will be presented by: