Testing for canonical form orientation in speech tempo perception

Abstract

We report on two experiments that aimed to test the hypothesis that English listeners orient to full pronunciation forms-"canonical forms"-in judging the tempo of speech that features deletions. If listeners orient to canonical forms, this should mean that the perceived tempo of speech containing deletions is highly relative to the speech's articulation rate calculated on the basis of surface phone strings. We used controlled stimuli to test this hypothesis. We created sentences with one ambiguous word form (for example, support~sport), to give half of the listeners an orthographic form that includes support and the other half an otherwise identical orthographic form with sport. In both experiments, listeners judged the tempo of the sentences, which allowed us to assess whether the difference in imposed interpretation had an impact on perceived tempo. Experiment 1 used a tempo rating task in which listeners evaluated the tempo of experimental stimuli relative to comparison stimuli, on a continuous scale. Experiment 2 used a tempo comparison task in which listeners judged whether second members of stimulus pairs were slower or faster than first members. Both experiments revealed the predicted effect of the imposed word interpretation: sentences with an imposed "schwa" interpretation for the ambiguous word form were judged faster than (the same) sentences with an imposed "no schwa" interpretation. However, in both experiments the effect was small and variables related to the experimental design had significant effects on responses. We discuss the results' implications for our understanding of speech tempo perception.

Keywords: English; Speech tempo; articulation rate; deletion; perception.

Figure 1.

Visual interface for one experimental trial (Experiment 1).

Figure 2.

Distribution of Rating (N = 3,640).

Figure 3.

Estimated marginal means, with confidence intervals, for the combined variable levels of Block and Orthography in the optimal model for Rating (see Table 3).

Figure 4.

Visual interface for one experimental trial (Experiment 2).

Figure 5.

Estimated probabilities, with confidence intervals, for the combined variable levels of Order and Orthography in the optimal model for Response (see Table 7).