An experimental and computational study of an Estonian single-subject word naming experiment
This study reports a large-scale single-subject experiment combining the word naming task with eye-tracking. Response variables (first fixation duration, total fixation duration, number of fixations, word naming latency, and spoken word duration) were subjected to statistical analyses with generalized additive models (GAMs). We compared GAMs with classical lexical predictors such as word frequency and inflectional paradigm size with GAMs that were given predictors derived from computational implementations of the discriminative lexicon model (DLM) for the lexical route in the word naming task, using word embeddings to represent words' meanings. With the exception of word length, lexical measures and DLM-based measures were found not to predict initial fixation duration and total number of fixations. DLM-based measures were statistically well supported as predictors of total fixation duration, naming latency, and spoken word duration. GAMs with classical predictors provided more precise fits for the first two response variables, but a GAM with DLM-based predictors provided a superior fit to the spoken word durations. These results support the possibility that in the naming task, word meaning is involved. The GAMs for spoken word duration and response latency revealed a novel phenomenon, namely that when the lexical route is not functioning, due to the mapping from meaning to form being unsuccessful, response latencies increase, spoken word durations decrease, and the speaker likely resorted to naming with the non-lexical grapheme-to-phoneme naming route.