Hearing “words” without words: speech prosody and word perception

Abstract

The role of prosody in word perception was studied by asking listeners to locate word boundaries in speech with natural prosody but without phonotactic or semantic cues. To obtain such speech, seven speakers read sentences (e.g., “The noisy dog never stopped barking”) with “ma” substituted for each syllable of the three-syllable adjective-noun phrase (e.g., “The mama ma never stopped barking”). These nonsense phrases with all possible stress patterns were excised and played to 38 naive listeners who judged whether they heard each phrase as “ma mama” or “mama ma.” The stress pattern determined how well the phrases were parsed into words. When the stress pattern made the word division unambiguous (01-1, 1-12, 1-10 where 0, 2, 1 denote no stress, midstress, and high stress), listeners used the pattern to parse the phrases easily; striking individual differences were observed. When the stress pattern left the word division ambiguous (11-1, 1-11, 12-1, 1-21, 10-1, 1-01), parsing was harder but still better than chance. So not only global patterns, but also detailed information about prosody, are used in word perception.