ISCA Archive SpeechProsody 2014
ISCA Archive SpeechProsody 2014

Silent reading and prosodic structure constraints

Philippe Martin

Silent reading of written texts involves necessarily a process of subvocalization, i.e. the presence of a voice reading the text in the head of the reader speaking to her/himself. This process includes not only the sequences of syllables corresponding to the written material, but also sentence intonation. Since subvocalization cannot be eliminated other than by changing the status of each word into a pictographic function (as it may be the case for a STOP road panel sign), it is argued here that sentence intonation is essential to language comprehension, and more specifically to the conversion of sequences of syllables into higher order linguistic units (corresponding to accent phrases AP in the Autosegmental-Metrical model). Consequently, reading and in particular silent reading is constrained by the same rules than the prosodic structure in general, and specifically to the minimal duration of accent phrases. This minimal value, occurring when AP’s contain only one syllable, is about 250 ms, a value which corresponds to the minimal period value of Delta brain waves. Therefore this AP minimal duration limits also the maximal number of AP that could be processed in silent reading, i.e. about 240 per minute, which corresponds to the maximal number of words per minutes experts in fast reading can process while keeping a reasonable level of comprehension, i.e. about 800 wpm.