• Keine Ergebnisse gefunden

EXPERIMENT 6: STRUCTURAL DIFFERENCES III 143 Table 7.14: Fixed Effects of Reaction Times of Experiment 6

The Role of Structure 7

7.4. EXPERIMENT 6: STRUCTURAL DIFFERENCES III 143 Table 7.14: Fixed Effects of Reaction Times of Experiment 6

Estimate SE t-value HPD.lower HPD.upper Pr(>|t|) (Intercept) 1142.17 25.96 44.0 1100.01 1186.21 < .001 ***

Length -6.42 7.94 -0.8 -21.77 9.22 n.s.

Statusc1 -66.13 9.71 -6.8 -85.08 -47.13 < .001 ***

Statusc2 111.71 10.34 10.8 92.45 132.88 < .001 ***

Order -18.37 7.91 -2.3 -33.42 -2.65 < .05 *

Response 56.22 11.15 5.0 32.93 77.36 < .001 ***

Length×Statusc1 -9.41 19.27 -0.5 -46.09 28.73 n.s.

Length×Statusc2 -41.78 19.32 -2.2 -79.33 -2.50 < .05 * Length×Order 64.43 15.77 4.1 33.35 94.58 < .001 ***

Statusc1×Order -67.35 19.29 -3.5 -104.73 -29.26 < .001 ***

Statusc2×Order 77.01 19.49 4.0 37.47 114.84 < .001 ***

Length×Statusc1×Order -116.71 38.55 -3.0 -191.47 -40.87 < .05 * Length×Statusc2×Order -65.61 38.60 -1.7 -143.09 8.16 n.s.

Judgment Times. Experiment 6 reveals significant main effects of the factors Status, Order and Response for judgment times. The main effect of Status indi-cates that it took longer to judge ambiguous sentences (1126 msec) than to judge grammatical sentences (1056 msec; Status-c1: t = -6.8; p < .001; Estimate = -66.13). Grammatical sentences are judged faster than ungrammatical sentences (1183 msec; Status-c2: t = 10.8; p< .001; Estimate = 111.71). The main effect of Order reveals that acc-initial sentences are judged faster (1109 msec) than nom-initial sentences (1134 msec; t = -2.3; p < .001; Estimate = -18.37). The main effect of Response reveals that on average, subjects reacted faster when judging correctly (1105 msec) than when misjudging grammaticality of sentences (1187; t

= 5.0; p< .001; Estimate = 56.22). The factor Length does not reveal a significant main effect. The main effects have to be modified by interactions.

(i) Length×Status-c2. It takes significantly longer to judge short ungrammatical sentences than long ungrammatical sentences. Grammatical sentences, however, are not affected by the factor Length (t = -2.2; p< .05; Estimate = - 41.78).

(ii) Length×Order. The interaction (t = 4.1; p< .001; Estimate = 64.43) reveals that Length affects reaction times of acc-initial sentences. Short acc-initial sen-tences are judged faster (1094 msec) than long acc-initial sensen-tences (1123 msec).

Nom-initial sentences are not affected (short: 1152 msec; 1116 msec).

(iii) Status × Order. Both Status-c1 and Status-c2 interact with Order. Both interactions Status-c1×Order (t = -3.5; p< .001; Estimate = -67.35) and Status-c2×Order (t = 4.0; p< .001; Estimate = 77.01) show that grammatical acc-initial sentences are judged faster than grammatical nom-initial sentences.

(iv) Length × Status-c1 × Order. The significant three-way interaction reveals that long acc-initial ambiguous sentences are judged somehow slower than short initial ambiguous sentences. There is no difference in reaction times for acc-initial grammatical or nom-acc-initial sentences (t = -3; p< .05; Estimate = -116.71).

7.4.3 Discussion

First, short sentences are judged significantly better than long sentences. Thus, Length evoked the effect as predicted by Gibson (2000) which clearly reveals a locality effect in Experiment 6. This contradicts with assumptions of expectation-based hypotheses that predict a beneficial effect for additional length. The results also provide a significant interaction of Status×Order. The interaction shows that acc-initial ambiguous sentences reveal higher error rates than nom-initial ambigu-ous sentences. This interaction replicates previambigu-ous findings on ambiguity resolu-tion preferences (cf. Bader and Bayer, 2006:89 and literature cited therein) that are assigned in The Case Assignment Generalization: ‘assign nominative!’ (Bader and Bayer, 2006:89). Those findings reveal a clear garden-path effect. Further-more, ungrammatical sentences in Experiment 6 are always created as double-case violations (‘NOM-NOM’, ‘ACC-ACC’). The experimental data support previous findings (e.g. Schlesewsky et al., 2003) implicating that ungrammatical double nominative constructions are incorrectly judged as grammatical more often than double accusative constructions. Thus, the interaction Status×Order shows that, both syntactic ambiguity resolution towards the non-preferred analysis and the detection of case violations, result in higher error rates as already reported in pre-vious studies. However, the results for judgments of Experiment 6 failed to yield the suggested interactions with Length. Additional length does not seem to signif-icantly affect critical conditions more strongly than all other sentence conditions.

The results for correct judgments (cf. results for correct judgments in Figure 7.1) show that additional length adversely affects ungrammatical and acc-initial am-biguous sentences numerically. Probably due to a lack of statistical power, those tendencies fail to turn significant. Nevertheless, those tendencies clearly indicate that the critical conditions might be more sensitive to additional length than other sentence conditions. This, however, indicates that the loss of an items’ activation might result in the need to reaccess previously processed features. Reaccessing grammatical features seems to be more difficult when more costly distances are located between related items. Thus, the study reveals the tendency that syntac-tic ambiguity resolution and detection of case violation are more error-prone over longer distances.

Although results of judgments fail to provide significant statistical support for the hypothesis that critical conditions are more sensitive to length, findings in judgment times reveal some interesting data. The experiment yields significant

7.4. EXPERIMENT 6:STRUCTURAL DIFFERENCES III 145 results of the main factors Status (both contrasts: ‘Status-c1’ and ‘Status-c2’), Order and Response. The factor Length fails significance. This might be due to different patterns of the various sentence conditions. Numerically, nom-initial sentences indicate a slight advantage in reaction times for long sentences. The same pattern can be found in initial ungrammatical sentences. However, acc-initial ambiguous and acc-acc-initial grammatical sentences show a disadvantage for long sentences (cf. results for judgments times in Figure 7.1). Ungrammatical and acc-initial ambiguous sentences are judged the slowest which might reflect the additional time it takes to reanalyze the sentence. Those findings support earlier results of increased reaction times for both, ungrammatical sentences and garden-path sentences (cf. Bader and Bayer, 2006).

All effects have to be qualified by interactions. Grammatical acc-initial sen-tences are judged faster than grammatical nom-initial sensen-tences. Furthermore, judgment times reveal significant two-way interactions with Length. First, the re-sults show that it takes longer to judge short ungrammatical acc-initial sentences than long ungrammatical acc-initial sentences. This suggests a beneficial effect of length for detection of case violations. The three-way interaction shows that short acc-initial ambiguous sentences are judged faster than long acc-initial ambiguous sentences. This meets assumptions that ambiguous sentences that disambiguate towards the non-preferred acc-initial analysis are in a disadvantage and are thus judged slower than ambiguous sentences that disambiguate towards the preferred reading. Results for judgment times of Experiment 6 reveal mixed findings. It takes less time to judge sentences where case violations have to be detected when those sentences are long. This suggests a facilitating effect of length for detec-tion of case violadetec-tions. This might be the case because ungrammatical sentences are often not identified and thus judged faster. Second, it takes longer to judge garden-path sentences when they contain additional costly material.

In summary, results of Experiment 6 show a clear tendency: a higher sensi-tivity of critical conditions for additional costly length. Here, critical sentence conditions are defined as the tasks of (i) detection of case violations and (ii) of syntactic ambiguity resolution. However, the data fail to reveal a significant inter-action for this analysis. It seems, that the minimal number of only one intervening NDR significantly affected processing performance in all sentence conditions, but does not suffice to result in a significant interaction. However, the data support locality-based assumptions that an increased distance between related items ad-versely affects the parsing process. Furthermore, the results contradict assump-tions that suggest a beneficial effect with increased length. Findings in judgment times support previous experimental findings. It takes longer to judge ungrammat-ical and acc-initial ambiguous sentences. In addition, the findings partially back the assumption that the parser is sensitive as to the strength of locality effects. It takes longer to judge garden-path sentences when they are long. However, length

has a positive effect on ungrammatical sentences. This contradicts locality-based assumptions and reveals a certain speed-up in judging ungrammatical sentences as assumed in anti-locality hypotheses. There is a tendency in the results that sug-gests that, given a certain minimal threshold of costly intervening material, tasks such as garden-path recovery and the detection of case violations are more likely to be affected by length. As suggested in locality-based hypotheses: reaccessing of grammatical features gets more difficult when more costly material intervenes between the initial activation of information and the later point of its retrieval.

However, the findings in judgment times indicate that some cognitive tasks such garden-path recovery take longer to be judged and thus are more error-prone with additional length, while the detection of case violations benefits from a longer distance of related items.

7.5 General Discussion

Chapter 7 presented three experiments that investigated working memory perfor-mance in German wh-constructions. Experiments 4 and 5 tested the effects of different embedded syntactic structures on sentence processing performance. To this end, German wh-clauses were manipulated with the presence or absence of (i) relative clauses and (ii) adverbial clauses. Crossing the presence and absence of both clause types results in four specifications (Short, RC, ADV, Long) for the fac-tor Length. Crossing Length with the facfac-tor Order (nom-initial, acc-initial) yields an experimental 2×4 design with eight sentence conditions. Both Experiments 4 and 5 were run using the moving-window self-paced reading procedure (cf. Just, Carpenter, and Wolley, 1982) that allows to analyze participants’ reading times on a word-by-word basis. After an initial analysis, single words were subsumed into various regions. Analysis of the experimental results of Experiments 5 revealed valuable findings.

First, adding either one of the additional clauses results in increased processing complexity. Findings in the regions following the respective lengthening (either AUX or NP2), reveal increased reading times compared to sentence conditions that have not been prolonged with additional material. This spillover effect argues in favor of locality-based assumptions. Expectation-based hypotheses predict ben-eficial effects of the additional material. This was supposed to manifest in faster reading times in the adjacent regions following both the relative clause or the ad-verbial clause.

Second, the region AdvV shows an effect of Order that reveals faster reading times for acc-initial sentences compared to nom-initial sentences. In order to form an acc-initial grammatical sentence, there are still missing (a) a second NP and (b) the matrix verb. In contrast, in nom-initial sentences only the matrix verb is

7.5. GENERAL DISCUSSION 147