3.2 B EHAVIOURAL EXPERIMENTS : T HE PRESENT TENSE OF G ERMAN STRONG VERBS
3.2.4 Crossmodal immediate repetition priming (experiment 2)
Proponents of the crossmodal priming technique (Clahsen et al., 2001a) claim that it taps into the mental lexicon more effectively than unimodal techniques and that it is very accurate with respect to timing (Tabossi, 1996). Many studies used this technique in order to gain empirical support for particular models of lexical access and representation.
The priming prediction for the immediate crossmodal priming task are summarised below (Table 35). As in experiment 1, FUL expects no differences between verb classes. In contrast to experiment one, though, mixed verbs should prime equally well, independent of the root vowel [a] versus [e]. The disappearance of the mixed [a]-verb effect in experiment 1 presumably correlates with the concomitant semantic priming if the target immediately follows the prime. Other theories (exemplified in Table 27) would not predict a difference for mixed [a]-verbs between experiment 1 and 2. In FUL, on the other hand, a slight priming advantage for weak [a]-verbs is expected, since only in this condition, there would be a full match relation between the 2ND SG and the corresponding infinitive.
Table 35: Priming predictions for experiment 2.
CONDITION VERB CLASS ROOT VOWEL PRIME TARGET PRIMING GLOSS.
control - - - - no -
semantic - - - - yes -
morphological strong a schläfst schlafen yes sleep
morphological strong e siehst sehen yes see
morphological weak a machst machen yes (most) make
morphological weak e lebst leben yes live
morphological mixed a bäckst backen yes bake
morphological mixed e quillst quellen yes soak, well
3.2.4.1 Material
The same test stimuli as in experiment 1 were used. However, the crossmodal priming design required a different organisation of the prime-target pairs. In this experiment, the auditory primes were 2ND SG PRES of German verbs. They stemmed from the spoken word material of experiment 1. The visual targets were always infinitives.
Three group lists in a Latin-square design were constructed (as in experiment 1), so that no subject saw a target more than once. For each group list, 340 filler pairs were added to the 60 test pairs, 200 of which the visual target was a pseudoword constructed in the same way as described in experiment 1. The primes were always real words, and there were as many words as pseudowords in the group lists. The filler pairs were distributed such that each test pair was preceded by 3-7 filler pairs. The total amount of
items was 800 per subject (400 prime-target pairs) and the whole experiment was designed to last about 18 minutes. A training session consisted of 10 prime-target pairs which did not occur in the main experimental session.
3.2.4.2 Subjects and procedure
50 students took part in the experiment (mean age 26, 27 females). They participated for a reimbursement or for class credits. Subjects were required to make a word/nonword decision on the visually presented targets by pressing the appropriate word/nonword button of the same reaction-time measuring device as used in experiment 1. They were told not to pay attention to the acoustic stimuli, but to concentrate on the visual targets only. The reaction times were recorded from the onset of the visual targets. Each prime-target pair was preceded by an attention tone of 200 ms, followed by a 200 ms pause. The acoustic primes were immediately followed by the visual primes which were displayed for 200 ms on the subjectsÊ screens. The font type was Arial (36 mm). The stimuli were displayed at a distance of 60 cm from the subjects, so that no eye movement was necessary to read the items.
3.2.4.3 Results
Errors. In total, there were 6.4% wrong responses and 5.5% timeouts. Among the targets, 4.6% were wrong responses and 3.2% were timeouts. No subject or item was excluded. Outliers had reaction times
<200 ms or >1000 ms.
Table 36: Lexical decision times (in milliseconds) per condition and verb class (experiment 2). The right-most column shows the percentage of wrong responses to the
corresponding targets.
CONDITION VERB CLASS
LATENNCY LSM [ms]
STANDARD ERROR
WRONG RESPONSES
Control overall 593.22 10.03 3.9%
strong 577.73 11.11 2.4%
weak 609.00 11.21 4.8%
mixed 592.95 11.20 4.5%
Semantic overall 567.49 10.02 3.4%
strong 555.74 11.08 2.5%
weak 576.20 11.17 4.0%
mixed 570.50 11.18 3.6%
Morphological overall 557.59 10.01 2.2%
strong 551.9 11.11 1.8%
weak 562.13 11.13 2.8%
mixed 558.74 11.11 2.0%
- 125 - ANOVAs. Again, two ANOVAs were calculated:
The accuracy model with the factors SUBJECT, TARGET (nested under VOWEL and VERB CLASS),
CONDITION, VOWEL and VERB CLASS showed that the response accuracy differed between conditions (CONDITION: F[2,2782]=4.00, p<0.02) and verb classes (VERB CLASS: F[2,2782]=3.76, p<0.03). Subjects were less accurate in the control and in the semantic condition, and on mixed and weak verbs. As before, however, the interaction of CONDITION and VERB CLASS was not significant (F[4,2782]=0.41, p>0.8); neither was the three-way interaction of CONDITION, VOWEL and VERB CLASS (F[4,2782]=1.78, p>0.12).
The reaction time model with the same factors showed that CONDITION was significant (F[2,2627]=28.42, p<0.001, cf. Table 36). Planned comparisons showed that both semantic and morphological priming was significant (morphological priming: t=5.28, p<0.001; semantic priming:
t=7.31, p<0.001) but the magnitude of morphological priming was significantly higher compared to the magnitude of semantic priming (t=2.04, p<0.05). Priming was again indepent of the verb class (CONDITION X VERB CLASS: F[4,2627]=0.82, p>0.50, cf. Figure 17). Furthermore, the interaction
CONDITION X VOWEL X VERB CLASS was not significant F[4,2627]=1.77, p>0.13). Within the morphological condition, all verbs showed priming, independently of vowel and verb class, as revealed by further post-hoc tests (strong [a]-verbs: t=2.06, p<0.04; strong [e]-verbs: t=2.33, p<0.02; weak [a]-verbs:
t=5.31, p<0.001; weak [e]-verbs: t=2.17, p<0.04; mixed [a]-verbs: t=2.11, p<0.04; mixed [e]-verbs: t=3.73, p<0.002, see Figure 18).
Morphological Priming Experiment 2
0 10 20 30 40 50 60 70 80
overall strong weak mixed
Facilitation (Control-Test) [ms]
Morphological* Semantic*
Figure 17: Facilitation effects per verb class (experiment 2). Times are given as least square means in milliseconds, control-semantic (white bars) and control-morphological (black bars). Standard errors
are indicated on top of each bar (+/-). Significance is marked by an asterisk.
Morphological Priming per Vowel and Verb Class Experiment 2
0 10 20 30 40 50 60 70 80
a* e* a* e* a* e*
strong weak mixed
Facilitation (Control-Test) [ms]
Morphological Matching Scores (x40)
Figure 18: Facilitation effects per vowel and verb class (experiment 2). Times are given as least square means in milliseconds and represent the difference between the corresponding control and
morphological conditions (black bars). Matching scores (2NDSG PRES to INFINITIVES) are multiplied by 40 (white bars); for the mixed [a]-verbs, the matching score is the mean of the nomismatch (0.25) and the mismatch (0) relation. Weak [a]-verbs showed significantly more priming than weak [e]-verbs. Standard errors are indicated on top of each bar (+/-). Significance is
marked by an asterisk.
As before, a correlation between morphological priming and matching score (2ND-infinitive) was calculated. This time, the correlation was significant (R=0.26, F[1,59]=4.26, p<0.05). Infinitive facilitation thus crucially depended on feature matches and nomismatches. More priming was obtained if the matching score approached 1. Further planned comparisons showed that priming significantly differed between the root vowels [a] and [e] in the weak verbs (24 ms priming advantage for [a]-verbs, t=8.48, p<0.005, see Figure 18). There were no vowel based differences in the mixed (t=-0.76, p>0.44) or in the strong verbs (t=0.10, p>0.91).
Comparing unimodal and crossmodal 2ND SGÆINFINITIVE priming. In order to directly compare the outcome of experiment 1 and 2, a further ANOVA was calculated, additionally comprising
EXPERIMENT (unimodal, crossmodal) as between variable. Crucially, there was no interaction of
CONDITION, VERB CLASS and EXPERIMENT (F[4,6691]=0.25, p>0.92). Hence, the overall morphological effects of experiment 1 could be replicated in experiment 2 and were thus independent of the experimental design (see Figure 19). On the other hand, the differences between experiment 1 and 2 based on the absence versus the presence of semantic priming were indicated in the significant interaction of CONDITION and EXPERIMENT (F[2,6691]=4.43, p<0.02).
- 127 -
2nd SG-Infinitive - Comparison (Experiment 1 & 2)
-10 0 10 20 30 40 50 60 70 80
strong weak mixed strong weak mixed
Unimodal Crossmodal
Facilitation (Control-Test) [ms]
Morphological Semantic
Figure 19: Comparison of facilitation effects per verb class in experiment 1 and 2. Times are given as least square means in milliseconds, control-semantic (white bars) and control-morphological
(black bars).
3.2.4.4 Discussion (experiment 2)
The most intriguing result of experiment 2 is that except for the semantic priming and the priming in the mixed [a]-verbs, the results of experiment 1 could be replicated: Morphological priming was significant in all three verb classes, and did not differ significantly between the classes. As predicted, semantic priming occurred, but was less than the combination of morphological and semantic priming.
The reaction time facilitation was also independent of the root vowel, but altogether, the amount of priming correlated significantly with the featural matches between the 2ND SG forms and their respective infinitives. Priming in the mixed [a]-verbs was now significant, possibly due to the concomitant semantic effects which overrode the subtle morphophonological phenomenon of experiment 1. Thus far, the predictions have been borne out successfully.
The comparison with experiment 1 showed that the general morphological effects were robust and did not depend on the experimental technique. This provides further evidence that priming depends on featural matches, nomismatches and mismatches in the relation of a 2ND SGPRES to its corresponding infinitive. The facilitation of the infinitive does not depend on whether it is presented auditorily (experiment 1) or visually (experiment 2). Both presentation modalities connect to the same core lexical representation, as illustrated in the previous chapter.
At first sight, the results are compatible with the predictions of full listing models. On the other hand, the correlation of priming amount and feature matching55 does not follow readily from such
55 Note however that this correlation was again not very strong. The objectives of the experiments in this dissertation were not to find behavioural correlates of matches versus nomismatches. For that purpose, matches
theories. This is not surprising, though, since full listing models generally do not look at featural representations. Thus, there should not be any priming differences based on the quality of the root vowel.
The comparison of experiment 1 and 2 showed that the auditorily presented 2ND SG PRES primes could activate their respective infinitives independently of the target modality. In experiment 1, targets were presented auditorily, while in experiment 2, subjects were presented the infinitive forms visually.
Both modes of presentation led to the same behavioural results. This strongly suggests that in both cases, the primes pre-activated their corresponding lexical (underlying and modality-independent) representations. Once these representations were active, it did not matter whether the target was perceived auditorily or visually: The lexical decision on the target forms was sped up compared to the control condition in which no pre-activation took place. Hence, the priming in experiment 1 and 2 tapped into „truly‰ lexical representations and was brought about through genuine morphophonological relations in the lexicon. No intermediate representations have to be assumed with respect to the outcome of the two experiments. Once again, the results supported the FUL model which assumes feature-based lexical core representations onto which both auditory and visual speech information is mapped.
Again, these results are at odds with traditional dual route models which relate differences in priming to verb class differences. Dissociations between weak [a]- and [e]-verbs would not be predicted.
The dual route model of Clahsen and colleagues, in contrast, makes interesting predictions regarding the prime-target direction. In this framework, as shown above, a priming difference is expected between pairs like schläfst-schlafen and schlafen-schläfst. The 2ND SGÆINFINITIVE direction ought to prime more effectively than the INFINITIVEÆ2ND SG direction, since in the former case, all morphosyntactic features of the target are already present in the prime. This does not hold for the latter case, and consequently, priming should be less effectively. These predictions were tested in experiment 3 and 4 which reversed the prime-target directions of experiment 1 and 2 in the respective modalities.