F 0 and segmental length contrasts

One of the novel aspects of this thesis was to investigate two types of prosodic con-trasts within one study; nonnative segmental length concon-trasts and pitch concon-trasts. Both types of contrasts are absent in the German lexicon. However, pitch contrasts exist in the German intonation system. The production experiments (Experiment 1 and Exper-iment 4) tested the two prosodic properties together within an utterance/nonce-words and the perception experiments examined the opposite process, selective attention to differences in one of the properties. To test this in the discrimination tasks, one of the two prosodic cues was used as a task-irrelevant one for the discrimination of the other cue. In this section, I will propose two different underlying mechanisms for the process-ing of segmental length and pitch in “input” and “mental representations” and discuss the non-reciprocal effect of the task-irrelevant pitch on the discrimination of segmental length contrasts and the effect of the task-irrelevant segmental length structure on the discrimination of pitch contrasts.

First, given that the phonetic details of contrasts were still available in the short ISI condition, it was assumed that nonnative listeners would exploit the phonetic informa-tion to successfully discriminate nonnative prosodic contrasts in that condiinforma-tion. Such an advantage due to the phonetic information in the short ISI condition compared to the long ISI condition was found only in discriminating nonnative segmental length con-trasts, but not pitch contrasts: Experiment 2 that tested the discrimination ability of non-native segmental length contrasts showed that even the learners and the non-learners reached d’ scores as high as those of the Japanese in the flat pitch and the short ISI con-dition. However, their d’ scores decreased in the long ISI concon-dition. Experiment 3 that tested the discrimination ability of pitch contrasts showed that the learners and non-learners reached lower d’ scores than those of the Japanese already in the short ISI

con-6.1 General discussions 131 dition. These divergent findings between the segmental length and pitch contrasts pose the question, why the phonetic advantage was not found for the German participants to discriminate pitch contrasts when the phonetic details of the contrasts were still avail-able.

One could also assume that the acoustic correlates of the segmental length contrasts were just more perceptually “extreme” than the pitch contrasts for the German listen-ers. However, the acoustic measurement of the stimuli showed that the pitch falls had in average over 16 semitones, which was a large pitch range. Previous studies showed that pitch ranges of declarative sentences in German were around 9 semitones (8.1 semitones with a sd = 2.3 semitones for female speakers and 9.7 semitones with a sd = 2.9 semitones for male speakers in Zimmerer et al., 2015 and 7.7 semitones with a sd = 2.6 semitones for female speakers and 10.2 semitones with a sd = 0.57 semitones for male speakers in Brinckmann and Benzmüller, 1999). Though it is difficult to compare the perceived salience of the segmental length contrasts with that of the pitch contrasts, one could as-sume that both contrasts were perceptually salient enough to obtain the phonetic effect.

Further, there are other possible explanations:

It may be the case that a listener is phonetically more sensitive to the acoustic cor-relates of a contrast of a prosodic cue that has the least linguistic constraint in one’s L1 (Hallé et al., 2004). German L1 listeners can be assumed to be phonetically more sensitive to nonnative segmental length contrasts than to pitch contrasts, because a pitch contrast is linguistically (post-lexically and paralinguistically) used in German, while a consonant length contrast is not. Since consonant length contrasts are not present in the mental lex-icon of the German listeners, they were phonetically more sensitive to acoustic correlates of consonant length contrasts. Pitch contrasts convey linguistically meaningful contrasts in German, and therefore the German listeners may have been less sensitive to acoustic correlates of pitch contrasts in comparison to those of consonant length contrasts, be-cause the absence of the phonological contrast may enhance the listener’s sensitivity to the acoustic correlates of a stimulus. Hallé et al. (2004) that tested a cross-linguistic Chi-nese tone discrimination ( of /pa/, /pi/ and /kwo/ produced in the four ChiChi-nese tones) reported that French L1 listeners had a psychoacoustic advantage in discriminating pitch contrasts over English L1 listeners and even a more considerable one over L1 listeners of tone languages, because French listeners are constrained the least byF₀ in their L1 among these three groups. Mandarin Chinese employs lexicalF₀and English usesF₀to contrast lexical stress and post-lexical information (Beckman, 1986). French usesF0to mark segmental and prosodic boundaries (Delattre, 1951) and French intonation is

char-132 General discussion and outlook acterised by a sequence of rising pitch movements, which limits a dynamic use ofF₀to convey post-lexical and paralinguistic information as English does.

Another possible interpretation for the asymmetric findings for the discrimination of segmental length contrasts and pitch contrasts is supported by the claim that tone lan-guage listeners established fine-grained associations between pitch contours and lexi-cal meaning and this presumably leads to a generally high ability in language-unspecific pitch processing (Deutsch et al., 2006; Pfordresher and Brown, 2009). This could be true also for Japanese L1 listeners, whose L1 employs lexical pitch contrasts. Based on this claim, a listener may be phonetically more sensitive when a prosodic cue conveys lexical contrasts than when it does not. Applying this claim to German, German listeners may be more sensitive to segmental length contrasts than to pitch contrasts for the following rea-sons: Although pitch is used contrastively in German, pitch contrasts are not associated with lexicon and intonation patterns do not relate to any specific stressed syllables of a word, so that pitch information may be not stored together with a word. As for segmental length contrasts, German listeners may have mental representations of segmental length contrast assigned to specific words in their L1 (e.g. Stadt vs. Staat). This explains why the German listeners were phonetically more sensitive to the segmental length contrasts than to the pitch contrast in my experiments.

Furthermore, I examined whether a task-irrelevant prosodic cue affected the discrim-ination of the other prosodic cue and whether the effects were reciprocal. The results of Experiment 2 showed overall a strong effect of the task-irrelevant pitch on the discrimi-nation of the nonnative segmental length contrast. The nonnative listeners’ d’ scores and the RTs were constantly affected by the task-irrelevant pitch, indicating that the Japanese lexical pitch movement (e.g. a falling lexical pitch accent or an initial low, cf. Haraguchi, 1977) makes the perception of the nonnative consonant length contrast more difficult when it occurs simultaneously. As for the contrary effects, an effect of the task-irrelevant segmental length structures on the discrimination of the pitch contrast was found only in the learners’ and non-learners’ RTs. The learners’ RTs were longer for the stimulus pairs presented in the more German-like segmental length structure than for the stim-ulus pairs presented in the less German-like segmental length structure. In the long ISI condition, both the learners and the non-learners slowed down to a greater extent in the less German-like segmental length condition than in the more German-like segmental length condition. The task-irrelevant segmental length structure did not affect the sensi-tivity to the pitch contrasts itself (= d’ scores), but the task difficulties (= RTs). In Chapter 4, I argued that the more German-like segmental length structure of the stimuli activated

6.1 General discussions 133 the nonnative listeners’ L1 phonological representations to a greater extent than the less German-like segmental length structures and this probably led to differences in their RTs.

I postulate that the effects of the task-irrelevant prosodic dimensions on the discrim-ination of the task relevant prosodic contrasts related to what extent words in the mental lexicon were activated. The consistent effects of the task-irrelevant prosodic dimension found in both the d’ scores and the RTs for the segmental length contrasts are evidence that the discrimination of segmental length contrasts activated words in nonnative lis-teners’ mental lexicon to the greatest extent, followed by the discrimination of pitch con-trasts presented in the long-vowel structure, and finally by the discrimination of pitch contrasts presented in the singleton and geminate structure. As discussed before, the pitch fall presented in the long-vowel structure was phonologically more natural to the German participants than the one presented in the singleton or geminate structure. This led to the greater extent of word activation in the mental lexicon.

Further, there are several arguments to why I hypothesise that the attention to the tar-get segmental length contrasts activated words in the mental lexicon to a greater extent than the attention to the target pitch contrasts did. In order to discriminate segmental length contrasts, participants paid attention to that target cue. While paying attention to segmental length contrasts, segmental features (Hall, 2007) to identify vowels and con-sonants were inevitably processed together. By activating the vowels and concon-sonants of the stimuli, words in the mental lexicon were activated. Note that pseudo-words were used as stimuli, so that the activation of words in the mental lexicon could have taken place to a lesser degree than when using real words. What is more, it was an initial acti-vation without a final selection of words. On the contrary, pitch is not used for the lexical distinction in German, but to code paralinguistic information such as the attitude and emotion of a speaker. Therefore, it can be hypothesised that the German listeners did not activate words while paying attention to the pitch contrasts so much as while paying attention to the segmental length contrasts. As mentioned above, pitch patterns are as-sumed not to be stored together with words (= there may be no one-to-one relationships between words and pitch patterns) in the German listeners’ mental representations. An-other argument would be that to discriminate pitch contrasts, segmental features can be ignored much more easily than when discriminating segmental length contrasts. The stimuli could have been replaced with sine wave sounds or Gaussian noise or with any other tonal sounds like music in order to examine the pitch contrasts. However, to exam-ine segmental length contrasts, vowels and consonants are necessary to build vowel and

134 General discussion and outlook consonant length contrasts. Otherwise it would have become a discrimination task for rhythmic contrasts and not for vowel and consonant length contrasts.

To conclude, different underlying processing mechanisms for the different prosodic cues (e.g.F₀and segmental length) were discussed in this section. Based on such under-lying processing differences, “prosody” should not be regarded as a collective (umbrella) term and treated as such, as if there would exist a general processing of “prosody”, but should rather be investigated separately when talking about e.g. segmental length and pitch. The discussion further highlighted the uniqueness of prosody; the same cue is used multiply at different linguistic levels (at lexical, post-lexical and paralinguistic lev-els) in different ways in each language. Indeed, lateralisation and neuronal representa-tions of the same prosodic cue differ depending on one’s L1 and L2s (e.g. Barry, 1981;

Näätänen et al., 2007; Tamaoka et al., 2014; Zhao et al., 2011).