The Brazilian Portuguese eye tracking corpus with a predictability study focusing on lexical and partial prediction

Abstract

Our main objective with this thesis is to present the first corpus of eye tracking measures in reading with predictability indexes in Brazilian Portuguese. Predictability is understood, in general, as an ability we have to predict words before they appear in text or speech (Staub, 2015; Kuperberg; Jaeger, 2016). Currently, there is evidence that linguistic predictability could act at different levels (partial predictability), and not being restricted to an exact word prediction (lexical predictability), as in the prediction of grammatical classes, or number and gender inflections (Kuperberg; Jaeger, 2016; Luke; Christianson, 2016). In our research, we studied the predictability of 2494 words, in a total of 50 short, self-contained paragraphs, divided into three genres: literary, journalistic, and pop science. In all, data from 286 participants who answered the Cloze task and 37 other who participated in the eye tracking reading task were analyzed. We found evidence that lexical prediction is a rare phenomenon and that only 4.8% of content words were highly predictable on average, while 10.8% of function words were highly predictable on average. However, we found that 33% of the content words had high rates of partial predictability, and the same occurred in 19% of the function words. Our analyses of the Cloze task answers showed that shorter words that are at the end of sentences were produced more quickly. We found evidence that predictable words were also produced more quickly. The analyses of eye movement showed that both lexical and partial predictability facilitated processing in reading, but the influence of lexical prediction, although it occurs more rarely, was more pronounced. Our analyses, therefore, showed that predictability was influential for language production and comprehension, being facilitative for both.