Proceedings Fonetik 2009 - Institutionen för lingvistik

Proceedings, FONETIK 2009, Dept. of Linguistics, Stockholm Universityfinal > initial (p initial(p=0.19)).Finally, for the initial word, the interactionof FOCUS and SPEECH ACT was not significantfor word duration (cf. Table 1). That is,[vaˈlandəɹ] was produced longer in assertionsthan in confirmations, both when in focus and inpre-focal position (cf. also Figure 2). Post-hoctests for FOCUS show that [vaˈlandəɹ] is realisedwith a longer duration when the word is in focusthan when focus is on the medial (p=.011) orfinal word (p=.003). However, when only thestressed syllable is taken into account, the interactionof SPEECH ACT and FOCUS is significant(cf. Table 1). As shown by post-hoc comparisons,the situation is, however, more complexthan for the interactions found for the otherwords: First, [land] is realised longer in assertionsthan in confirmations not only when theinitial word is in focus (p=.002), but also whenthe final word is in focus (p=.029). Second, inassertions, the duration of [land] differs in allthree focus conditions (initial focus > medialfocus (p=.015); initial > final (p=.036); final >medial (p=.039)), while in confirmations, [land]is significantly longer in focus than in the twopre-focal conditions only (initial > medial(p=.005); initial > final (p=.016)), i.e. no significantdifference is found between the twopre-focal conditions.In the analysis so far, all recordings havebeen included irrespective of the variation of F0patterns produced within an experimental condition.As mentioned in the Introduction, confirmationswere produced with either of twostrategies, as classified in Ambrazaitis (in press)as either ‘rising’ (presence of a (lowered) H-accent on the target word), or ‘falling’ (absenceof a H- accent on the target word), cf. Figure 1.This raises the question as to whether the focallengthening found in confirmations (cf. Figure2) is present in both the rising and the fallingvariants. Figure 3 displays the results for confirmationin a rearranged form, where the F0pattern is taken into account.For the medial word, Figure 3 indicates that,first, the word seems to be lengthened in focuseven when it is produced with a falling pattern(cf. “förlänger” in conditions ‘medial fall’ vs.‘final fall’, ‘final rise’, and ‘initial rise’), andsecond, the focal lengthening effect still tends tobe stronger when the word is produced with arise (‘medial fall’ vs. ‘medial rise’). However,for the final word, focal lengthening seems to bepresent only when the word is produced with arise. Finally, the initial word seems to bems550500450400350300250initial rise(38)Wallander förlänger novembermedial fall(17)medialrise (20)final fall(16)final rise(27)Figure 3. Mean durations of the three test words inconfirmations, divided into classes according to theintended focus position (initial, medial, final word)and F0 pattern produced on the target word (rise,fall); n in parentheses.lengthened not only when it is in focus itself, butalso when medial or final focus is produced witha fall, as compared to medial or final focusproduced with a rise.DiscussionThe goal of this study was to examine the durationalpatterns in a data corpus where focus waselicited in two different speech acts, assertionsand confirmations. It is unclear from the previousF0 analysis (cf. Figure 1 and Ambrazaitis, inpress) whether focus was successfully signalledin confirmations, when these were producedwithout a ‘rising focal accent’ (H-). The generalhypothesis to be tested in future perception experimentsis that focus in confirmations mayeven be signalled by a falling pattern, whichwould support the proposal by Ambrazaitis(2007) that there is a paradigmatic utterancelevelaccent contrast in Standard Swedish betweena rising (H-) and a falling accent.The present results are in line with this generalhypothesis, since they have shown that focallengthening can be found not only in assertions,but also in confirmations, although the degree offocal lengthening seems to be smaller in confirmationsthan in assertions. In fact, the speechact hardly affects the duration of unfocussedwords, meaning that speech act signalling interactswith focus signalling. Most importantly,the results also indicate that focal lengtheningmay even be found when the target word isproduced with a falling F0 pattern, although noinferential statistics have been reported for thiscase. In fact, in these cases, duration differencesseem to the more salient than F0 differences (cf.‘medial fall’ and ‘final fall’ in Figures 1 and 3).76