Using SNF instead of SPT affected the creation of the ambiguous stimuli to be used as the basis for further manipulations. Unlike SPTs, SNF have a peak aligned earlier than QNF, a phonetic property mirrored in the different analyses and transcriptions of Narrow Focus accents in questions (L*+H, see Section 2.1.3) and statements (L+H*). Averaging the peak alignment required a 2 ms manipulation for E1 and a 15 ms manipulation for E2. However, informal testing from three NI native speakers confirmed that the resulting stimuli sounded both ambiguous and natural, thus qualifying as viable bases for further manipulations. The use of two base stimuli instead of one did not lead to an increase in the number of experimental stimuli because instead of manipulating four (sets of) cues we only manipulated one. For E2 there were 14 test items (2 base stimuli x 1 dimension x 7 steps), i.e. half of those used in E1 (1 base stimulus x 4 dimensions x 7 steps). This allowed us to increase from 18 to 34 the number of control natural items. Control items were composed by the two unresynthesized base stimuli and by two repetitions of 16 trisyllabic Subjects extracted from the Tre Grazie corpus. This time, we excerpted the names from 4 utterances of 2 different sentences in the 2 pragmatic contexts. Again, the first names used for control items were different from the one used for test items, but in this case the percentage of the three first names was perfectly balanced.43, whereas the test name was almost four times as frequent as each of the other two in E1.44 Subjects participating to E1 spontaneously reported a certain degree of sensitivity to the relative frequency of the test items’ name (see Section 3.2.5). Thus, for E2 stimuli we rescaled the relative frequencies of the three names. Moreover, after the test we asked subjects whether they considered one of the three names to be frequent than the others, but none reported any noticeable skew. As for the experimental items, we modified the curve index by shifting the height of rise midpoint in 7 steps, identical to the procedure for the second set of E1, using the PSOLA algorithm (Moulines & Charpentier 1990) embedded in Praat (Boersma & Weenink 2008). This was done for both ambiguous base stimuli. The central step (n. 4) was assigned a value that corresponded to a linear interpolation between L1 and H; steps from 3 to 1 had progressively higher height values, corresponding to a progressively increasing concave interpolation, and steps from 5 to 7 had progressively lower height values, corresponding to a more and more convex interpolation (see Figure 3.6). Step size was 15 Hz, as determined through the use of actual values of rise midpoint in the two base stimuli (used as penultimate in both direction) and the number of steps.

3.3.2 Hypotheses

Experiment

Results show that subjects do not use L1 alignment, rise shape or L2 scaling as cues for the classification of trisyllabic stimuli as narrow focus questions or partial topic statements. The interpretation of negative results being an epistemologically complex operation, in the following we will concentrate on some hypotheses accounting for this outcome, and on further testing needed in order to validate them.

The resulting stimulus was used as the starting point for further f0 manipulations. Four sets of stimuli were created by manipulating three dimensions and one of their combinations (see Figure 3.4 and Table 3.1). The three dimensions were L1 alignment, rise shape and L2 scaling, whereas the fourth set combined rise shape and L2 scaling manipulation. Each of the 4 sets was composed by 7 steps which were equally spaced in frequency as expressed in Hz.39 Values of the ambiguous base stimulus were assigned to the central step (n. 4); values of the original stimuli were assigned to penultimate steps in the two directions (i.e. SPT = 2 and QNF = 6). This allowed us to create 4 sets which, for each parameter, went from an overtly SPT characterized stimulus (n. 1) to an overt QNF (n. 7).

in the sense of useful information not yet encoded in abstract accounts of intonation. In order to test this hypothesis, we devised a

Figure 3.2: Phonological analyses and f0 contours of the five naturally produced stimuli Katherina sucht ’ne Wohnung; (1)-(3) were produced as statements, (4)-(5) were produced as questions. The syllables -ri- and Woh- that showed the prenuclear and nuclear accents are delimited by horizontal lines. Readapted from Petrone and Niebuhr (2014), Fig. 2.

Studies on the role of dynamic f0 contour cues in perception bring together these two lines of research, in that they rest at the same time on a deep understanding of how much detail in the signal is psychoacoustically perceptible and of how intonational categories can be identified. In this section we will review two such studies, dealing with NI (Petrone & D’Imperio 2011) and with Northern Standard German (Petrone & Niebuhr 2014), respectively. Both studies analyze prenuclear falls and show that sentence modality contrasts are not exclusively cued by the intonational nucleus.

yield

al u

track

Kontaktlinsen für jedes Auge speziell angefertigt werden.

rs fü

insen

n grau