• Keine Ergebnisse gefunden

3 Regular Stem Allomorphy

3.1 Overview

3.1.1 Trisyllabic shortening

3.1.1.3 Discussion: Trisyllabic shortening

The goal of the first two experiments was to determine how regular stem allomorphs are listed in the English mental lexicon. We hypothesized that regular stem allomorphs share a lexical entry. To test our hypothesis, the error detection mechanisms were studied in three violation conditions, viz. ser[i:]nity (RD),

*ser[aɪ]nity (UD), and *seromity (NC), in comparison with the baseline condition serenity (W). We designed two word list experiments with different tasks: a lexical decision task and a memory task. We argued that the RD items could be structurally repaired if they share the stem morpheme with the W items. In this case, the error-detection mechanisms elicited by the RD items should be different from those, evoked by the irreparable nonwords (UD and NC conditions).

However, if regular stem allomorphs were listed separately, all violation conditions should have elicited similar brain responses. The results of the lexical decision task experiment revealed a graded N400 effect of the type W<RD<UD=NC. The change of the experimental task resulted in a different set of violation responses. Thus, in the early processing stages the RD und UD items elicited PMN effects, while in the late processing stages only the RD condition evoked LPN.

The first experiment employed a lexical decision task to specifically engage the semantic processing. We expected that the error-detection mechanism triggered by the semantic processing would result in the N400 effect.

Indeed, the lexical decision task yielded the N400 that was most pronounced for the irreparable items (UD and NC). The RD condition elicited a mild N400 effect.

These results were in line with our predictions that the RD items should evoke a mild N400 effect if they can be structurally repaired. Such structural repair could only take place if there was a possibility of accessing the underlying stem morpheme with the subsequent application of the (morphophonological) TSS rule.

The purpose of the implementation of the lexical decision task was two-fold: (1) the elicitation of the semantic processing, and (2) the possibility of structural reanalysis. Based on the first purpose, our expectations were to observe a differential N400 effect for the violation conditions. We adopted the morphological decomposition as a prerequisite for the storage of allomorphs in

the mental lexicon. Considering the obligatory decomposition, the RD nonwords were first decomposed into their constituent morphemes and were then recomposed with the combined meanings of those morpheme constituents. The whole procedure would thus be indexed in the attenuated amplitude of the N400 effect. McKinnon et al. (2003) have already employed the N400 as an index of the morphological decomposition. They argued that if all complex words were decomposed prelexically, the nonwords made up of bound morphemes (*intain,

*retrude) would elicit a similar violation effect as low-frequency bound-stem complex words (intrude, retain) would. The authors reasoned that a nonword made up of existing though non-productive morphemes should be treated as a low-frequency word, hence the above-mentioned argumentation. The results of the study supported the authors’ hypothesis, having revealed a similar mild N400 effect for both bound-stem conditions. In contrast to McKinnon et al. (2003), Koester et al. (2007) claimed that the N400 reflected the process of semantic composition. In an experiment on German compound nouns, they compared the processing of the semantically transparent compounds with the processing of the semantically opaque ones. As only low-frequency semantically transparent compounds evoked an N400-like negativity, the authors deduced that the N400 effect reflected the integration process of the semantics of the morpheme constituents into the compound.

McKinnon et al. (2003) used non-productive morphemes in their study, while Koester et al. (2007) employed a productive means of word formation in German, viz. compounding. These two studies are hardly comparable due to the discrepancies in the structure of the stimulus materials, nevertheless they provide supporting factors for the present study. The lexical decision task experiment employed in the present study differed from McKinnon et al.’s (2003) and Koester et al.’s (2007) studies in the implementation of a productive suffix {-ity} and an existing stem that is never found in the environment of this suffix. More importantly, in order to combine with the suffix {-ity}, this stem has to undergo TSS. This structure of the critical items placed them in an intermediate position in comparison with the stimulation used in the previously mentioned studies. On the one hand, the RD items were zero-frequent, their constituents, on the other hand, were relatively high-frequent morphemes. Though never found in the English vocabulary, the RD nonwords could still be semantically recomposed by accessing the meaning of the constituent morphemes. In line with the arguments put forward by McKinnon et al. (2003) and Koester et al. (2007), the RD items should evoke a reduced N400 effect reflecting both the morphological

decomposition and semantic composition processes. The results of the lexical decision task experiment provided evidence for our assumption that semantically transparent complex words are subject to prelexical morphological decomposition.

The second goal of the lexical decision task was devised exclusively for the purpose of ensuring the reanalysis procedure. We were interested in the manner of representation of the TSS rule. If this rule is not included into the lexical entry and is therefore listed in a separate morphophonological layer, the violation thereof must trigger a grave morphosyntactic conflict. According to O’Rourke & Van Petten (2011), the P600 amplitude could be modified by a grave morphological violation. Van de Meerendonk et al. (2008) also claimed that only a grave conflict between the expected and the actual input could trigger the P600 effect. The gravity of the morphological violation or the lack thereof in the RD condition could be explained by the following two factors. First, Chomsky & Halle (1968) put forth the idea that the ability of a stem vowel to undergo the TSS should be listed in the lexical entry and not in a separate morphophonological layer. Following their line of argumentation, the violation of the TSS rule should be reflected in the semantic processing and not in the syntactic one, hence the absence of the P600 effect in the lexical decision task experiment. The second factor is the depth of derivation. The W items underwent the TSS simultaneously with the attachment of the suffix {-ity}, which could describe this kind of word-formation as direct or one-step derivation. The only conflicting input in the RD items was the violation of the expected laxing of the stem vowel. According to the present results, the gravity of this conflict was not sufficient for the induction of the P600 effect.

Summarized, the results of the lexical decision task experiment delivered evidence for the morphological decomposition/ semantic composition of the semantically transparent complex words. Additionally, they provided support for the consolidation of the morphophonological layer into the lexical entry. However, we maintain at this point that this is true, at least, for the trisyllabic shortened stems.

The memory task was introduced in order to shift the focus from the lexical semantics onto the general structure of the stimuli. Considering the fact that the items were presented auditorily in the test block but the probe for the actual task was presented visually, the subjects were not only to memorize the items but also had to re-encode them in another modality. This task required both linguistic processing and the involvement of the short term and episodic memory.

We predicted a pronounced PMN effect for the RD items and a reduced PMN effect for the irreparable nonwords during the initial processing and a distinct LPN for the RD nonwords during the post-retrieval stage. The results of the study provided supporting evidence for our expectations: the PMN effect was mild for the UD and NC items and was most prominent for the RD condition. Only RD items evoked the LPN effect.

The first finding of this study demonstrated the effects of the morphophonological context. Taking into account the homogeneity of the stimulus materials, the expected phonetic form of the stressed stem vowel should be characterized by the TSS, i.e. these vowels should surface as short vowels with a changed height. The RD condition was made up exclusively by violating the TSS rule, yielding tense vowels in the stressed position. The conflict between the expected input and the actual acoustic signal reached its apex during the presentation of the RD items as all of them deviated from the contextual constraints. The irreparable violation conditions consisted of both the trisyllabic shortened vowels and tense vowels in the proportion 50:50. The PMN effect was similarly reduced for these conditions demonstrating an intermediate amplitude value lying between the W and the RD conditions. Though the occurrence of the PMN effect in no manner reflected the organization of the mental lexicon entry, it showed that the amount of deviation from the expected context played an important role at the early processing stages (Connolly & Phillips, 1994; Connolly et al., 1992; Connolly et al., 1990; D'Arcy et al., 2004; Steinhauer & Connolly, 2008). However, we observed PMN exclusively in the memory task experiment.

The absence of the PMN effect in the lexical decision task might be explained by the overriding N400 component that could have concealed the initial phonological conflict.

The initial processing stages reflect automatic processes that cannot provide reliable information about the structure of the lexical entry. The stages that are important for the study of the mental lexicon are the retrieval and the post-retrieval stages. During the lexical retrieval stage (300-500 msec post stimulus) no violation effects were observed. The lack of the violation effects in this particular time window could be explained by either the prior PMN effect, which could have reduced possible lexical effects, or by the task demands. The participants did not have to perform a lexical decision task, which distracted them from the lexical semantics. With the semantics out of the picture, the only violation type that required high maintenance cost during the post-retrieval stage was the formal violation, viz. the RD condition. The analysis of the late time

window (900-1100 msec) revealed a significant LPN effect for the RD items. The remaining question was, however, how the induction of the LPN effect was possible without the implementation of a recognition paradigm. The paradigm that we employed in the memory task experiment did not involve direct source monitoring. Nevertheless, though different from the standard recognition paradigm, our task still implicated the maintenance of contextual information. On the one hand, the phonetic structure of the stimulus had to be memorized, and on the other hand, the possible ways of conversion of this structure into the orthographic form had to be computed. The items that contained conflicting information increased the context retrieval demands. The effort for the processing of the RD items had to be doubled as has been argued above. First, the stem morpheme had an incorrect surface form that could never be combined with the suffix {-ity}. In order to combine with the suffix, the stem vowel had to be trisyllabic shortened. Therefore, the deviation from the standard phonetic structure required effort from the episodic memory to retrieve the information concerning the occurrence of the stem allomorphs. Second, though completely different in the surface form, the two allomorphs had the same orthographic code.

The effort required for the computation of the orthographic form for the RD items contributed to the increase of processing cost. The influence of these two factors resulted in a non-canonical but robust LPN effect.

It could be argued whether the LPN reflected the processing within or without the lexical entry. The arguments against the processing within a shared lexical entry may be the lack of violation effects during the retrieval stage and the non-linguistic nature of the LPN effect. Though valid, such an argument could not explain why irreparable nonwords behaved in the same way the existing words did. During the retrieval and the post-retrieval stages all conditions except the RD items elicited similar brain responses. The surface form of the W items met the requirements of the morphophonological context; besides, these items were existing words. The only necessary effort for the processing of the W items consisted in keeping them in the short term memory. The irreparable items, despite having a faulty phonetic form in half of the cases, did not have a stem that could be retrieved from the mental lexicon. According to the absence of semantic violation responses, these items were accommodated into the episodic memory with both the phonetic and a possible orthographic form, and were simply kept ready for the memory task. Therefore, the lack of necessity (W) or the possibility (UD, NC) to maintain the stem attributes resulted in a normal/ standard processing of the existing words and the irreparable nonwords.

Contrary to that, the maintenance of the stem attributes was necessary for the processing of the RD nonwords. This procedure would be impossible without access to the lexical entry. We assume that this lexical entry is shared by the allomorphs employed in the W and the RD conditions. Otherwise, the violation of morphological combinability should have evoked the structural reanalysis. The impossibility of repairing the surface structure via the offline application of the TSS rule within the mental representation would have caused a strong morphosyntactic conflict indexed by a prominent P600 effect. This was not the case in the present study. The LPN effect elicited by the RD items indicated the access to a single underlying stem morpheme for the regular stem allomorphs.

In sum, the results of the word list experiments run with British participants provided supporting evidence for our hypothesis that regular stem allomorphs share a mental lexicon entry. This conclusion was deduced from the fact that the RD nonwords elicited violation effects that made this condition significantly different from the real word and both irreparable nonword conditions.

We posit that these effects were induced by the possibility of structurally repairing the RD items and not by such factors as the frequency of occurrence or the neighborhood size.