Listening material, instrumentation and procedure 45

The stimuli employed in Experiment 2 were constructed according the same main principles as those in Experiment 1. The same Finnish speaking female who produced the original stimulus for Experiment 1 recorded the pseudoword atta in exactly the same way as she did in the first experiment.

A set of ten stimuli were constructed from the original atta word (see Fig. 6). The whole set of stimuli formed an artificial continuum in which the duration of a stop closure was varied in stepwise fashion. This time, the continuum was constructed by deleting 20 milliseconds of silence at a time from the middle of the silent occlusion of the pseudoword atta. The

durations of the sounds were the following: the duration of the first vowel was 84 milliseconds, the occlusion of the word medial stop was 260 milliseconds and the following vowel was 103 milliseconds. The total duration of the original word was 447 milliseconds, and that of the whole set of stimuli varied from 267 to 447 milliseconds. The segmentation conventions for the phonetic durations were the same as in Experiment 1 (see pages 33-34). The Figure 5 illustrates the stimuli continuum with the different durations of occlusions. Further, the instrumentation and the procedure used in Experiment 2 were identical to that described in Experiment l.

0. 260 Al: 0.178

w.·

i.'� i

. ,,,11, •.

A2:0.434

·. '

I'

I

1i,I:;· ·j� ^I �

I

I _� .. ,,fl,y

�rn·

·T

100 150 200 250 300 350 400 450 500 550 600 msec 4500 Hz 4000 3500 3000 2500 2000 1500

■

I I I

FIGURES Waveform and spectrogram representation of the naturally produced pseudoword atta illustrating the segmentation conventions employed in Experiment 2. The graphical illustration shows that the word final vowel was emphasised. This emphatic stress at the second syllable of atta ^is similar to that in the original pseudoword ata in Experiment 1 (see Fig. 2).

occlusion inmsec

FIGURE6

80 100 120 140 160 180 200 220 240 260

The stimuli continuum employed in the perception test of Experiment 2.

Duration of the silent closure stage of the word medial dental stop was reduced in stepwise fashion by increments of 20 msec. Natural speech sounds were used as stimuli. The original stimulus was taken from the speech of a female producing the pseudoword atta, which is stimulus 10 in the continuum.

Altogether 10 stimuli were constructed ranging in total duration from 267 to 447 msec, and the impression of the perceived stimulus shifted from atta ^to ata.

2.5.3 Results and discussion

All the stimuli were identified by all the subjects making 2100 identifications altogether. As in Experiment 1, the results indicate that the stimuli were identified in principle in the same way across the subjects. The following identifications were made: 97 percent (2039) of the stimuli were identified as either ata or atta, approximately 1,5 percent of the stimuli were attaa (32) identifications and another 1,5 percent were ataa identifications (29). These responses reveal that there was no ambiguity as to the quality of the sounds in this experiment.

The graphic representations of the responses in Fig. 7 demonstrate that the stimuli continuum was perceived categorically. The location of the category boundary of the stop was at approximately 125 msec. This crossover point was at a shorter duration than it was in Experiment 1. Obviously, differences in speech rate may have affected the differences in the category boundaries between the two experiments but since there was an attempt to control speech rate in the construction of the stimuli (the original stimuli were spoken within the same linguistic context by the same speaker) the rate of speech is not believed to have a significant effect on the results.

At the crossover point the proportional durations of the segments were the following: the proportional duration of the word initial vowel comprised 27 percent of the whole word, that of the word medial consonant 40 percent and that of the word final vowel 33 percent. In Experiment 1 the corresponding figures at the theoretical category boundary were the following: the proportional duration of the word initial vowel comprised approximately 21 percent of the whole word, that of the word medial consonant approximately 41 percent and that of the word final vowel approximately 39 percent of the whole word duration. These results suggest that when the proportional duration of the word medial consonant is 40

percent or more then it is categorised as long. In fact, the same tendency can be seen in Lehtonen's data (1970, 176) as well. When the proportional duration of the consonant in the original word miitii was increased and reached over 40 percent of the total duration of the word (42 percent and 45 percent) then the word medial consonant was categorised as long.

100 which the rest of the stimuli were constructed by shortening the occlusion of the stop by 20 msec at each step of the continuum. The category boundary is located near stimulus atta ^3.

Furthermore, the results could be interpreted as showing a reciprocal relationship between the proportional durations of the word medial consonant and that of the word final vowel. In both of the experiments it appears that when the proportional durations of the word medial consonants became larger than those of the word final vowels then the word medial consonant was categorised as a long consonant. In Experiment 2 the proportional durations of the C and V2 were 40 and 34 percent respectively at the crossover point, but when the same figures were of the order of 35 and 36 percent in the stimulus atta2 then the word medial consonant was categorised as short. In Experiment 1 the fi_gures in question were 41 and 39 percent at the crossover point but the stimulus ata2 with the figures 36 and 42 percent was categorised as a VCV word rather than a VCCV word. Therefore, it seems that both the proportional size of the word medial consonant and the relationship between it and that of the word final vowel affect the way in which the quantity of the word medial consonant in the VCV structures is categorised. Thus, the actual absolute duration of the

sounds does not seem to have a decisive role in the location of the category boundary between short and long quantity degrees. Furthermore, when the results of Experiment 1 and 2 are compared the assumption (e.g., Lehtonen 1970, 174) is confirmed according to which the longer the word final vowel is the longer the preceding single consonant can be before it is categorised as long.

The syntagmatic relationship between the word initial vowel and word final vowel does not seem to have any decisive role in the quantity categorisation of the word medial consonant. Since the physical durations of the vowels were not altered between the different stimuli, neither did the relation between the durational proportions substantially change between the word initial and word final vowels.

The bandwidth of the identification was approximately 57 msec. This is virtually the same as that in Experiment 1 (58 msec). However, the shape of the curves or the steepness of the category boundary between the two experiments appear to be different: the category boundary in Experiment 2 appears to be less steep than that in Experiment l. But since there was no difference in the bandwidth it cannot be said that the shape of the identification curves indicates that the stimuli employed in Experiment 1 were categorised more distinctly than those in Experiment 2. In this connection it should be noted, however, that the order of the experiments may have had an effect on the categorisation functions. When Experiment 2 was conducted the subjects were already familiar with this type of psychoacoustical experiment and with stimuli similar to those used in this experiment, and the listening strategies and expectations could have been different in the two experiments.

The identification responses of the individual subjects are presented in Figure 8. These responses revealed that the deviant responses as to the quantity category of the word final vowel, i.e., ataa and attaa responses, were traceable to the same individuals as in Experiment l. The male subject (Subject number 6 in Fig. 8) made the most deviant identifications: he identified the stimuli from the shorter end of the duration continuum in 28 of the cases as ataa, and in 20 cases in the stimuli from the longer end of the continuum as attaa. The ataa identifications were distributed in his responses in the following way: 10 times in response to stimulus 1, 9 times to stimulus 2, 7 times to stimulus 3 and twice to stimulus 4. The attaa identifications were distributed in the following way: 5 times in response to stimulus 5, 3 times to stimuli 6, 2 times to stimulus 7 and 8, and once to stimulus 9.

The female subject identified the shorter end of the continuum stimuli as ataa twice when stimulus 2 was presented and once when stimulus 3 was presented. The attaa identifications she gave in the following cases: 5 times when the stimuli 5 and 6 were presented, 3 times in response to stimulus 7, twice to stimulus 8 and once to stimulus 9. It appears that these two individuals seem to identify the stimuli in a deviant way to the rest of the subjects in both of the experiments. Thus, it seems that the way in which the stimuli continua were constructed did not affect to a

50

significant extent the way in which the quantity category of the word final vowel was identified by these two individuals nor the rest of the subjects. It could be that the adopted response strategies or the idiolects of these two speakers affected the way in which they responded in the two experiments.

100 ^{SUB ECTl}

FIGURE 8 Categorisation functions of individual subjects in Experiment 2. The black line shows the atta responses and the gray line the ata responses. The last graph in the right column shows the mean average of the responses.

100-.--,..,.-..i,..;..,..,,.,.

The result that in general the quantity categories of the adjacent sounds were not affected by the changes in the duration of the word medial consonant differs from the results of Lehtonen. In his study (1970, 184) the stimuli, in which the original long word medial consonant was shortened, were relatively systematically identified as having also a long word final vowel (miittiiii). No explanation for this discrepancy between the results of Lehtonen and the present study is readily available apart from the obvious fact that the stimuli were different. One explanation, however, could be the fact that he used real Finnish words as the original stimuli which also had real word counterparts (e.g., miitii 'rotten' and miittiiii 'to shove'), whereas in the present study the stimuli were not meaningful in Finnish. Therefore, the perceptual analysis of the stimuli in the present study had to rely more on the acoustics of the stimuli presented. In employing real words the knowledge of the possible alternatives of real Finnish words between the temporal changes of the segments in the stimuli continuum may have facilitated the variability in the responses in Lehtonen' s study.

The graphical presentation of the individual categorisation functions is presented in Fig. 8. These results indicate that most of the subjects located the categorical crossover point in between stimuli 2 and 3. As in Experiment 1, there does not seem to be any drastic difference in this function between the individuals. It should be noted, however, that the categorisation function between categories is not as steep as it was in the responses in Experiment 1. Perhaps the durational and proportional durations of the stimuli continuum were not as distinct as those in Experiment 1, and thus the stimuli were not as good exemplars of the word structures in questions as those in Experiment 1.

The categorical functions of the individual subjects as well as all the categorical functions of the two experiments presented in this Chapter demonstrate that there are no absolute durations for short and long quantity degrees of the consonants in VCV /VCCV structures. Nevertheless, the variable locations of the category boundary were still located within a relatively restricted durational stretch. It appears that some factor or factors affect/ s within which the crossover point of durational categories fall.

In

study by Nakajima, ten Hoopen, Hilkhuysen & Sasaki (1992) in which nonspeech sounds with similar stimuli durations to the present study were employed the point at which the listeners changed their categorical judgement from short to long was around the same as in the present study (120-160 msec). This evidence in addition to other evidence may be an indication that there is some kind of natural area in the durational continuum in which our perception changes from short to long. In fact, researchers of categorical perception have suggested that. category boundaries are based on natural sensitivities of the auditory system (e.g., Rosen & Howell 1987, 120-121). Thus, it is known that since there is an element of learning in categorical perception this learning may well be facilitated by some kind of natural perceptual sensitivity of the auditory system. Whether this perceptual sensitivity is the product of spoken communication of whether it is otherwise characteristic to humans or whether the same sensitivity is encountered in other species are questions

which should be answered in the future in order to understand more about temporal speech processing in general.

Also the fact that there seems to be a larger durational area in which the long quantity degree is located in comparison to the area of short quantity degree is evident from the results of the two experiments. The same tendency is also clearly evident in previous research on durational aspects of quantity distinction (e.g., Lehiste 1970, Lehtonen 1970, Magga 1984, Engstand & Krull 1994a, 1994b). Penttila (1963, 21) noted that in Finnish short sounds cannot be lengthened significantly without them becoming long whereas the durations of long sounds can be considerably extended without any inconvenience. Trubetskoy (cited in Lehiste 1970, 361) also suggested already in the 1930s that the short degree quantity degree seems to be restricted to a narrow point in time whereas the long degree has more of a length dimension which is stretchable at will. This suggestion gained more experimental support in the experiment described here.