Children's processing of inflected words

Clahsen et al. (2004) examined the production of regular and irregular participle forms of German with high and low frequencies using a speeded production task. Forty children in two age groups (5- to 7-year-olds, 11- to 12-year-olds) and 35 adult native speakers of German listened to stem forms of verbs presented in a sentential context and were asked to produce corresponding participle forms as quickly and accurately as possible. Dependent variables were the participants' participle-production latencies and error rates. Table 1 presents a summary of the main findings on the two groups of children. The scores in Table 1 display the percentages of error and the production latencies for high-frequency participle forms subtracted from those of low-frequency ones; cases in which these differences were significant are indicated by an asterisk. A positive value reflects an advantage, a negative one a disadvantage for high-frequency forms.

Children's production of participle forms revealed clear regular/irregular contrasts. Regular inflectional forms (i.e., the unmarked stem and the regular -t participle suffix) were overapplied to verbs that require irregular forms in the adult language (8.3 and 20.9%, respectively), whereas overapplications of irregular patterns to verbs that are regularly inflected in the adult language were extremely rare (less than 1%). Regulars and irregulars were also affected by frequency, but in different ways. The error rates shown in Table 1 reveal a significant frequency advantage for irregulars, with more errors on low-frequency irregular verbs than on high-frequency ones, but not for regulars. Contrasts between regulars and irregulars are also evident from the production latencies: high-frequency irregular participles were produced significantly faster than low-frequency ones, whereas regular participles yielded a reverse frequency effect, that is, significantly longer production latencies for high-frequency forms than for low-frequency ones.

The finding that high-frequency irregulars are produced more quickly than low-frequency ones and elicit fewer overregularization errors indicates that irregular participles have full-form memory representations in the children's lexicon. If an inflected word form is stored in the mental lexicon, then retrieval should be faster for high-frequency forms than for low-frequency ones, and this contrast should be reflected in different production latencies. Moreover, memory storage and retrieval are dependent on frequency of exposure, and hence, low-frequency forms should yield more errors than high-frequency ones. The results reported above are compatible with these expectations. By contrast, the suffixation errors (i.e., -t overregularizations) are the result of the child's applying a regular (-t) suffixation rule in cases in which the lexical entry for an irregular participle form is not available. Likewise, overapplications of the unmarked stem form to participles that require irregular stems arise when specific stem information is not available or not accessible to children. In such cases, they fall back on the base entry, producing errors such as *gefinden instead of the correct gefunden (found). In this way, the unmarked base stem serves as a default form in circumstances in which the required irregular forms are not retrieved. An explanation for the reverse frequency effect for regulars that was also seen in some previous studies with adults (Beck, 1997; Prasada, Pinker & Snyder, 1990) is offered by Pinker (1999). He argues that the production of an inflected word invokes both lexical lookup and the rule route, and that the lexical memory system is connected to the computational (“rule”) system by an inhibitory link. Thus, the activation of a stored whole-word representation turns off the rule, a mechanism that is independently required to account for the so-called blocking effect, that is, the fact that an irregular form blocks the application of the rule, for example, sang blocks *singed. Pinker claims that high-frequency regulars (but not low-frequency ones) have whole-word representations stored in memory. Thus, the production of high-frequency regulars involves memory access, which interferes with the rule route. This produces extra processing costs and slows down the production of high-frequency regulars relative to low-frequency ones for which the rule route is not impeded by any stored forms.

The study of children's speeded production of participles revealed regular/irregular contrasts parallel to those reported for adults, suggesting that the two mechanisms for morphological processing (lexical storage and morphological decomposition) are also employed by children. Children's production of participle forms was found to be different from adults' in three respects: higher overregularization rates, longer production latencies, and stronger and more consistent reverse frequency effects for regulars. These differences can be attributed to slower and less accurate lexical access and retrieval in children than in adults. Overregularization errors arise when access to the lexical entry of an irregular form fails (see Marcus et al., 1992). Consequently, we expect children to produce more such errors than adults. That children take longer to produce participles is also consistent with slower lexical access in children than in adults. Finally, reverse-frequency effects arise from the retrieval of stored high-frequency regulars that inhibit the rule route (Pinker, 1999). If lexical retrieval is relatively slow, then it will take longer to block the rule route, and thus the consistent reverse frequency effects in both groups of children.

In a second study we investigated children's on-line comprehension of German noun plurals using the ERP violation paradigm (Lück et al., 2001). Three age groups of children (6- to 7-year-olds, 8- to 9-year-olds, 11- to 12-year-olds) and an adult control group listened to sentences containing plural forms in two conditions: correct irregular or incorrect (overregularized) plurals (…Tuben/*Tubes…[tubes]), and correct/incorrect regular plural forms (e.g.,…Waggons/*Waggonen…[wagons]). ERPs were timelocked to the onset of the critical noun plurals and were averaged for a time window of 200–1500 ms. Recall that previous studies with adults yielded a left anterior negativity (LAN) for overregularized plural forms (relative to the correct ones) followed by a centroparietal positivity (P600), and an N400 for incorrect regulars. The question Lück et al. (2001) addressed was whether these components are also present in children.

Although plural overregularizations (condition 1) yielded consistent ERP effects in the children (see below), there was no N400 or any other significant difference between incorrect and correct regulars (condition 2) in either group of children, indicating that they were sensitive to overregularized plurals (*Tubes), but not to violations such as *Waggonen. The contrast observed in the ERP results was confirmed by an additional elicited production task in which the children were found to reliably produce the correct plural forms of the condition 1 items (e.g., Tube), with a mean accuracy score of 98%, whereas for the condition 2 items they performed at chance level (mean accuracy=51.8%). The children were not confident of the correct plural forms of the condition 2 items (perhaps because most of the items used in this condition were loan words), and as a result violations such as *Waggonen did not yield any measurable ERP effect. By contrast, the ERP results for condition 1 show that the children were sensitive to plural overregularizations. Table 2 indicates the distribution of the two ERP waveforms for correct plural forms subtracted from the corresponding overregularized ones.

Two groups of children (8- to 9-year-olds, 11- to 12-year-olds) showed a biphasic ERP waveform to incorrect noun plurals, that is, a frontal negativity followed by a centroparietal positivity, parallel to the pattern seen in adults. The timing of these two ERP components with the negativity having an earlier onset than the positivity was also parallel to the one for adults. In previous studies on adults, the anterior negativity was interpreted as signaling rule-based morphological processing. An -s overregularization such as *Tubes is word-internally decomposed into stem plus affix, and the item Tube is identified as a noun that takes an irregular plural form and therefore blocks the -s rule. The later positivity suggests that regularization errors cause additional processing at a sentence-level stage, perhaps because participants try to repair or reanalyze the regularized plural form before integrating it with the rest of the sentence. According to Lück et al.'s (2001) findings, effects of these processes are also seen in children above the age of 8. By contrast, the younger group of children displayed an unspecific broad negativity and no positivity. It is not quite clear what this ERP response might mean. One possibility would be that the younger children failed to identify the overregularized plurals as inflected forms of existing German words, perhaps because the -s plural rule was not yet completely stable and automatized. Independent evidence for that comes from the fact that in the elicited production task, the 6- to 7-year-olds achieved low accuracy scores for nouns requiring -s plurals (mean 29.4%). As a result, they may have perceived items such as *Tubes as unanalyzed wholes, in which case the negativity could be taken as an ERP response to a pronounceable nonword.

Adult L2 learners' processing of inflected words

There are four L2 studies employing speeded production or grammaticality judgment tasks to investigate the processing of inflected words (Beck, 1997; Birdsong & Flege, 2000; Brovetto & Ullman, 2001; Lalleman, van Santen, & van Heuven, 1997). The findings from these studies are inconsistent and partly surprising. Whereas the native speaker controls showed a consistent response-time advantage for high-frequency irregulars (but not for high-frequency regulars) in all experiments, most studies failed to replicate this contrast for the L2 learners. The L2 learners' results on regulars were even less conclusive (see Hahne et al., 2006, for discussion).

As the available behavioral studies have not been able to provide a clear picture of morphological processing in an L2, Hahne et al. (2006) used ERPs to investigate how L2 learners process inflected words on-line. Two inflectional systems of German, participle formation and noun plurals, were examined in a group of advanced L2 learners with Russian as L1 who had learned German after childhood. Hahne et al. (2006) adopted the designs and materials of previous ERP violation experiments with native speakers of German, from Penke et al. (1997) for participles and from Lück et al. (2001) for noun plurals. In both cases, participants were presented with sentences containing two types of violation: overregularizations and incorrect regulars. In addition to the ERP experiments, Hahne et al. (2006) performed two elicited production tasks with the L2 learners after the EEG sessions, in which participants were given uninflected verb and noun forms of the test items used in the ERP experiments and were asked to produce the corresponding participle and noun plural forms. Performance in these two tasks was good (>95% correct for participles, >86% for plurals) indicating that the participants were familiar with the critical items and their correct inflected forms.

Table 3 presents an overview of the ERP effects that Hahne et al. (2006) obtained for L2 learners and, for convenience, the corresponding results from adult native speakers of German. Table 3 illustrates that the group of advanced L2 learners studied by Hahne et al. (2006) responded differently to violations of regular and irregular inflection during on-line morphological processing. For misapplications of regular rules of inflection, they showed ERP effects that have independently been argued to tap morphosyntactic processing, namely an anterior negativity and/or a P600, whereas misapplications of irregular inflection revealed an ERP effect (the N400) that has been claimed to be characteristic of lexical processing and interpretation. The brain responses seen in L2 learners for these two kinds of morphological violation suggest that the two processing routes posited by dual-mechanism models of inflection (lexical storage and morphological decomposition) are also accessible and employed by L2 learners. Note, however, that although N400 and P600 effects were seen in both the native speakers and the L2 learners, early anterior negativities in the L1 speakers were more focal and consistent across experiments than in the L2 learners. Moreover, whereas the anterior negativity had its maximum at left frontal sites in the L1 speakers, it was observed bilaterally in Hahne et al.'s L2 group. Topographic variations of anterior negativities in this time range have also been observed in a number of ERP studies with adult native speakers, though; see, for example, Rodriguez-Fornells, Clahsen, Lleo, Zaake, and Münte (2001) for Catalan, Gross, Say, Kleingers, Münte, and Clahsen (1998) for Italian, and Weyerts, Penke, Münte, Heinze, and Clahsen (2002) for German, who found either bilateral negativities or even effects at right anterior sites. Both in terms of its timing and its distribution, the anterior negativity Hahne et al. (2006) found for the L2 learners falls within the range of variation that has been observed in studies with native speakers and can be taken to be an instance of a “morphosyntactic negativity,” an ERP waveform that is clearly different from the centroparietal N400, which has been found to be associated with lexical–semantic processing (see, e.g., Osterhout, 1997).

In native speakers of German, a LAN was found in three different experiments on participles (Penke et al., 1997) and in both the visual and the auditory ERP studies on noun plurals (Lück et al., 2001; Weyerts et al., 1997). In the L2 learners, an anterior negativity was found for participles but not for noun plurals. Given that the anterior negativity reflects early automatic processes of word-internal morphological decomposition, these results suggest that L2 learners employ these processes for participles but not for plurals. Proficiency differences are likely to be responsible for this contrast. An independent elicited production task with the L2 participants on the critical items used in the two ERP experiments yielded considerably worse correctness scores for plurals than for participles. For plurals, 14% of the elicited forms given by the L2 participants were incorrect, whereas there were hardly any errors for participles, indicating that the L2 learners were less confident in plural than in participle formation. Thus, it seems that the L2 learners are not only more confident in forming participles than noun plurals, their processing of participles is also more nativelike than that of plurals.

Preliminary summary

Clearly, more research on morphological processing in language learners is needed before any strong conclusions can be drawn. However, what the results from the above studies suggest is that child L1 learners (at least in the age range under study) do not fundamentally differ from adult native speakers in how they represent and process morphologically complex words. The two mechanisms that mature speakers have been shown to employ for morphological processing (lexical storage and morphological decomposition) are also used by children. Lexical storage effects were seen in the on-line production of irregular participles, and evidence for morphological decomposition comes from the speeded production and ERP studies. Differences between children's and mature speakers' processing of inflected words were argued to be due to slower lexical access and retrieval or to incomplete acquisition. The dual processing system, however, appears to be the same for children and adults.

In regard to L2 learners, the N400 effects seen for irregularizations are indicative of lexical storage of irregulars, parallel to what has been claimed for native speakers. This is compatible with previous studies investigating lexical–semantic processing in L2 learners and bilinguals in which N400 effects were obtained for semantic anomalies (Ardal et al., 1990; Hahne, 2001; Hahne & Friederici, 2001; Weber-Fox & Neville, 1996) in nonnative speakers that were similar to those found in monolingual studies. Likewise, the finding that P600 effects were seen in the L2 learners in the same conditions as for native speakers is compatible with previous ERP studies of syntactic processing in nonnative speakers, which demonstrated P600 effects for phrase-structure violations (Hahne, 2001; Weber-Fox & Neville, 1996) and violations of subject–verb agreement and gender concord (Sabourin, 2003). Finally, the fact that the learners' proficiency in the L2 seems to affect the ERP findings is also familiar from previous studies. For example, Friederici, Steinhauer, and Pfeifer (2002) report findings from an artificial grammar experiment in which adult subjects were trained on an artificial language system (BROCANTO) to a level at which they were highly proficient and produced hardly any errors. A subsequent ERP experiment examining syntactic violations in BROCANTO revealed the biphasic ERP pattern familiar from comparable studies of natural language processing in native speakers, that is, an early negativity followed by a P600. Taken together, these results suggest that at least in domains in which they are highly proficient, L2 learners can employ the same mechanisms for morphological processing as L1 speakers.

Children's processing of ambiguous sentences

Results from several studies of children's ambiguity resolution preferences suggest that children are more restricted than adults in their ability to make use of lexical–semantic and pragmatic information during parsing. Evidence for children's reduced ability to use lexical–semantic cues to disambiguation comes from a self-paced listening study by Felser et al. (2003). Felser et al. investigated how 6- to 7- and 10- to 11-year-old English-speaking children and adults processed relative clause attachment preferences in sentences such as The doctor recognized the nurse of (with) the pupil who was feeling very tired. Previous research on adult native speakers has shown that disambiguation preferences are affected by the type of preposition joining the two potential antecedent noun phrases (NPs). NP2 disambiguation (i.e., associating the relative clause with the pupil) is preferred cross-linguistically if the two possible antecedent NPs are joined by a thematic preposition such as with (Carreiras & Clifton, 1999; Gilboy, Sopena, Clifton, & Frazier, 1995; among others). For antecedent NPs joined by the case-assigning preposition of or its translation equivalents, on the other hand, attachment preferences have been found to vary across languages (Cuetos et al., 1996; Gilboy et al., 1995). One explanation for the robust NP2 preference for NPs linked by semantically contentful prepositions is that prepositions such as with create a local thematic domain of their own, and that the parser prefers to associate ambiguous modifiers with material inside local thematic domains (Frazier & Clifton, 1996). In the absence of such lexical biases, attachment preferences are determined by other factors including phrase structure based locality principles (Gibson, Pearlmutter, Canseco-Gonzales, & Hickock, 1996).

Figure 1 provides an overview of the different participant groups' mean reaction times to the disambiguating auxiliary in Felser et al.'s (2003) study. Only the adult group in Felser et al.'s (2003) study showed a significant interaction between preposition and attachment, indicating that their attachment preferences were influenced by the type of preposition involved. The children differed from the adult controls in that their disambiguation preferences were not affected by the type of preposition (of vs. with) at all. Instead, the younger children's on-line attachment preferences were found to interact with their working memory span as measured by Gaulin and Campbell's (1994) listening span test for children. Although both groups of high-span children showed a preference for NP1 attachment irrespective of the preposition involved, the low-span 6- to 7-year-olds showed an overall preference for NP2 disambiguation. These results suggest that depending on their listening span, the children applied one of two different phrase structure based locality principles during parsing. Whereas high-span children seemed to follow a “predicate proximity” strategy according to which ambiguous modifiers are attached as close as possible to the main predicate (compare Gibson et al., 1996), low-span children tended to associate the relative clause with the most recently processed NP, in accordance with the “recency” principle. Note that although the older children's reaction times were as fast as the adults' overall, the 10- to 11-year-olds showed the same nonadultlike pattern of preferences as did the younger high-span children, a preference for NP1 attachment irrespective of the type of preposition.¹

Overview of children's and adults' mean listening times (ms) on the disambiguating segment (Felser et al., 2003).

Results from a reading-time study by Traxler (2002) provide further evidence for children's tendency to choose the structurally simplest analysis, and show that they will do so even if this analysis is semantically inappropriate. Traxler examined 8- to 12-year-old children's processing of sentences containing temporary subject–object ambiguities such as When Sue tripped the girl (the table) fell over and the vase was broken, manipulating the plausibility of the ambiguous NP as a direct object of the preceding verb. The children tended to analyze the postverbal NP as a direct object regardless of whether or not this analysis was in fact plausible. A similar (albeit weaker) preference for the structurally simpler direct object analysis was observed even for intransitive verbs.

Results from a number of other studies have shown that children are less able than adults to take referential information into account during on-line ambiguity resolution, and provide further evidence for children's “least effort” approach to parsing. Trueswell et al. (1999) investigated 4- to 5-year-old children's processing of sentences containing temporary prepositional phase (PP)-attachment ambiguities such as Put the frog on the napkin in the box. Children's eye movements and actions were recorded as they responded to an experimenter's instructions to move objects around on a table. The visual context was manipulated (“one frog” vs. “two frog” contexts) so as to support either a verb phrase (VP) modifier or NP modifier reading of on the napkin. Whereas adults' attachment decisions were influenced by the type of context, in accordance with the referential principle (Altmann & Steedman, 1988; Crain & Steedman, 1985), the children interpreted the postverbal PP as the destination argument of the verb put in both one-referent and two-referent contexts. The fact that the children opted for the syntactically simpler VP-modifier analysis even in NP-modifier contexts suggests that children, unlike adults, are unable to integrate referential information provided by the visual context during on-line ambiguity resolution. The authors' observation that the children frequently performed inappropriate actions even after receiving disambiguating linguistic information moreover indicates that they are less able than adults to revise their initial analysis.

Results from a similar eye-movement study by Snedeker, Thorpe, and Trueswell (2001) confirmed children's inability to make use of referential information during sentence comprehension but indicated that 4- to 5-year-old children, like adults, were sensitive to verb bias when resolving PP-attachment ambiguities. That is, both children's eye movements and their actions in response to sentences like Examples 1a and 1b below revealed that they preferred the VP-modifier interpretation only for “instrument bias” verbs such as tickle in Example 1b but not if the stimulus sentences contained a “modifier bias” verb such as choose in Example 1a.

There is evidence, however, that children are not generally impervious to pragmatic information during language processing. Hurewitz, Brown-Schmidt, Thorpe, Gleitman, and Trueswell (2000) found that 4- to 5-year-old children were able to make use of the referential context in a production task, suggesting that their apparent inability to use contextual information may be task specific. The results from an act-out experiment reported in Meroni and Crain (2003), who used materials similar to Trueswell et al.'s (1999) but that involved two equally salient potential referents, also demonstrate that children are able to take referential information into account in off-line tasks. It is thus conceivable that only in situations of increased processing demand, such as during on-line comprehension tasks, do children prioritize on bottom-up information and disregard the referential context.

An eye-tracking study by Sekerina, Stromswold, and Hestvik (2004) on 4- to 7-year-olds' interpretation of ambiguous pronouns revealed a discrepancy between children's eye movements and their choices in a picture-selection task. Despite their eye movements indicating that the children were aware of a referential ambiguity, although they were slower to notice it than were the adults, they almost always selected a sentence-internal referent for ambiguous pronouns. Adults, on the other hand, chose a sentence-external referent in 20% of cases. As linking a pronoun to an external referent requires additional discourse-level processing, the authors suggest that the children's strong preference for a local antecedent might be due to their relatively limited processing capacity.

Taken together, the results from the above studies support the view that children apply the same kind of phrase structure based parsing heuristics as adults but are more limited in their ability to use lexical–semantic, plausibility, and referential information during on-line ambiguity resolution.

Ambiguity resolution in L2 processing

Results from several studies suggest that unlike children, L2 learners do not have any difficulty making use of lexical–semantic and pragmatic information when processing ambiguous input. How do L2 learners resolve syntactic ambiguities in the absence of any lexical or contextual cues for disambiguation though? The results from our studies in this domain indicate that L2 learners differ from both child and adult native speakers in that they do not rely on structure-based parsing strategies when resolving ambiguities in the L2.

Much of the work on L2 ambiguity resolution has focused on relative clause attachment ambiguities involving complex “genitive” (NP-of-NP) antecedents as in Someone saw the servant of the actress who was on the balcony. Given that relative clause attachment preferences are subject to cross-linguistic variation, investigating how L2 learners resolve RC attachment ambiguities may also help to shed some light on the “processing transfer” issue. Although mature monolingual English-speaking readers tend to prefer NP2 disambiguation (see, e.g., Carreiras & Clifton, 1999; Fernández, 2003; Roberts, 2003), NP1 attachment is preferred in many other languages including Spanish (Carreiras & Clifton, 1993; Cuetos & Mitchell, 1988; Gilboy et al., 1995), German (Hemforth, Konieczny, & Scheepers, 2000), French (Frenck-Mestre & Pynte, 1997; Zagar, Pynte, & Rativeau, 1997), and Greek (Papadopoulou & Clahsen, 2003). Several reading-time studies on a variety of L1/L2 combinations indicate that even highly proficient learners fail to acquire nativelike attachment preferences for relative clauses with complex genitive antecedents, consistently showing no preference for either NP1 or NP2 attachment instead (Dussias, 2003; Felser, Roberts et al., 2003; Papadopoulou & Clahsen, 2003; but see Frenck-Mestre, 2002).²

Papadopoulou and Clahsen's (2003) study examining advanced Spanish-, German-, and Russian-speaking learners of Greek moreover provides strong evidence against the transfer of L1 processing strategies. Despite the fact that the learners' L1s all pattern with Greek in that speakers of these languages typically exhibit an NP1 preference, none of the learner groups showed any attachment preference at all for relative clauses preceded by complex genitive antecedents in their L2. Parallel results were obtained in Felser, Roberts, et al.'s (2003) reading-time study on L2 English. Table 4 provides an overview of the results from Felser, Roberts, et al.'s and Papadopoulou and Clahsen's studies.

Together, these findings suggest that L2 learners are sensitive to the NP2 attachment cue provided by a thematic preposition but fail to show any clear attachment preferences for ambiguous relative clause modifiers when such lexical cues are absent, even if the preferred L2 ambiguity resolution strategy is the same as in their L1.³

Conceivably, the lack of any attachment preferences that has been observed in the above reading-time studies might be due to the learners' delaying their decision until the end of the sentence, or could even indicate that this method is inappropriate for measuring L2 parsing. The former possibility is unlikely given that the learners in Felser, Roberts, et al.'s (2003) and Papadopoulou and Clahsen's (2003) studies did not show any preferences for complex genitive antecedents in a complementary off-line task, either (but see Dussias, 2003). The latter possibility is ruled out by the authors' finding that all L2 groups tested exhibited a clear on-line preference for NP2 disambiguation if the two potential antecedent NPs were linked by a thematic preposition. Felser, Roberts, et al. (2003) and Papadopoulou and Clahsen (2003) suggested that the absence of any reliable preferences in the complex genitive conditions may be due to the learners' failing to apply phrase structure-based parsing principles (such as recency or predicate proximity; cf. Gibson et al., 1996) when processing ambiguous sentences in the L2, which they compensate for by overrelying on nonstructural cues to sentence interpretation instead. If such cues are absent, as in the genitive conditions, L2 learners' attachment decisions are made randomly.

Although the above studies have failed to find any evidence for L2 learners' use of phrase structure-based ambiguity resolution strategies, several other studies have demonstrated learners' sensitivity to lexical–semantic cues to disambiguation. Juffs (1998a), for example, found that L2 learners of English from various language backgrounds processed main verb/reduced relative clause ambiguities as in The bad boys criticized almost every day were playing in the park in a similar way as native speakers. The learners showed evidence of being “garden-pathed” if the initial participle looked like a transitive main verb (and if the following adverbial adjunct provided no cue as to the correct analysis), suggesting that the learners were sensitive to verb argument structure information during parsing. Learners from typologically different language backgrounds (Chinese, Japanese, and Korean speakers) had more difficulty processing ambiguous sentences of this kind than did learners whose L1 was typologically similar to English (i.e., Romance speakers) and were less accurate in judging the grammaticality of the test items.

In an eye-tracking study, Frenck-Mestre and Pynte (1997) examined how advanced English-speaking learners of French and French-speaking learners of English resolved PP-attachment ambiguities in sentences such as They accused the ambassador of espionage (of Indonesia) but nothing came of it. Both the learners and the native speakers associated plausible argument PPs such as of espionage with the VP and interpreted PPs such as of Indonesia as NP modifiers. Frenck-Mestre and Pynte also investigated learners' processing of sentences containing temporary subject/object ambiguities such as Every time the dog obeyed (barked) the pretty little girl showed her approval. Again, the learners behaved similarly to the native speakers in that sentences containing a potentially transitive verb elicited a garden-path effect compared to sentences containing intransitives. Both learner groups showed evidence of L1 lexical interference, however, in that they found sentences containing verbs like obey (which are intransitive in French but optionally transitive in English) more difficult to process than sentences containing verbs that are intransitive in both languages.

L2 learners' processing of subject/object ambiguities has also been investigated in self-paced reading experiments. Juffs and Harrington (1996), Juffs (1998b, 2004), and Felser and Roberts (2004) examined how L2 learners of English from various language backgrounds resolved temporary ambiguities in sentences such as After Bill drank the water proved to be poisoned. Again, the learners tested by Juffs and Harrington (1996) and Juffs (1998b, 2004) were garden-pathed in a similar way as the native speakers. Juffs (2004) furthermore observed that the size of the garden-path effect caused by sentences like the above was not correlated with the learners' working memory span as measured by the standard Daneman and Carpenter (1980) reading-span test (although there was some indication that word span might be a factor). Manipulating the plausibility of the postverbal NP as a direct object, Felser and Roberts (2004) found that advanced Greek-speaking learners of English were much more strongly influenced by plausibility information than native speakers when processing subject/object ambiguities in their L2, and that they had difficulty recovering from an initial misanalysis of sentences like the above.

Preliminary summary

The above findings reveal some interesting differences in the way children and adult L2 learners resolve structural ambiguities in real time. Results from a number of studies suggest that child L1 learners apply the same structure-based, least-effort processing principles as mature native speakers but are more constrained than adults in their ability to make use of nonstructural cues to interpretation.⁴

Children's processing of syntactic dependencies

Children have been shown to be sensitive to binding constraints in off-line tasks before the age of 5 (Chien & Wexler, 1990; Crain & McKee, 1986; Wexler & Chien, 1985). McKee, Nicol, and McDaniel (1993) used the cross-modal picture priming paradigm to examine whether 4- to 6-year-old children were capable of establishing syntactic binding relationships during the processing of sentences such as The alligator knows that the leopard with green eyes is patting himself on the head with a soft pillow. Like the adult controls, the children showed antecedent priming effects in the reflexive condition but not in the corresponding pronoun condition, demonstrating that they were able to apply their knowledge of binding constraints in real time.

Love and Swinney (1997) adopted McKee et al.'s (1993) cross-modal picture priming technique to investigate 4- to 6-year-old children's processing of filler-gap dependencies in sentences such as The zebra that the hippo had kissed________on the nose ran far away. Like adults (compare Love & Swinney, 1996), the children in Love and Swinney's (1997) study showed an antecedent reactivation effect at the gap position following the subcategorizing verb (i.e., kissed, in the example above). Note, however, that because the gap in Love and Swinney's experiment was located immediately after the verb, it is impossible to determine whether the observed priming effect reflected verb-driven (i.e., lexical–semantic) integration processes or structure-based gap filling.

Also using a cross-modal picture priming task, Roberts, Marinis, Felser, and Clahsen (2006) investigated how 5- to 7-year-old English-speaking children and adults processed sentences such as Example 2 below that involved three-place predicates.

Indirect object fillers were used in order to ensure that the putative gap was not directly adjacent to the subcategorizing verb. Participants were instructed to listen carefully to the stimulus sentences over headphones, and to make a lexical (“alive”/“not alive”) decision to pictures that appeared on a computer screen before them at different points during the sentence. Visual targets included pictures of the filler (e.g., a peacock) and pictures of unrelated objects such as a carrot. The picture targets were presented at two different test positions, at the gap position (e.g., after birthday present), or at a nongap control position 500 ms earlier. All participants additionally underwent a working memory test (Daneman & Carpenter, 1980, for adults, and Gaulin & Campbell, 1994, for children).

If the indirect object filler is mentally reactivated at the gap position, then lexical decision times should be faster for “identical” than for “unrelated” pictures presented at the gap position, and reaction times to “identical” pictures should be faster at the gap position than at the control position. Although the children's reaction times were slower overall than the adults', the results suggest that children's and adults' processing of filler-gap dependencies was affected by working memory differences rather than differences in lexical decision speed (compare also Nakano et al., 2002, for monolingual adults). For children and adults with a relatively high working memory span, a Position × Target Type interaction was found indicating that the filler (e.g., peacock in Example 2 above) was mentally reactivated at the gap position following the direct object NP. That is, high-span participants responded more quickly to “identical” than to “unrelated” picture targets at the gap position, and lexical decision times for pictures showing the filler were shorter at the gap position than at the earlier control position. Table 5 presents an overview of the high-span children and adults' mean reaction times to identical and related targets. Low-span children and adults, on the other hand, did not show any filler reactivation effects at the gap position. The results from the high-span participants tie in with Love and Swinney's (1997) earlier findings, and provide further evidence for phrase structure based gap-filling in L1 processing.

Short-term memory effects were also observed in Booth, MacWhinney, and Harasaki's (2000) study investigating 8- to 12-year-old children's processing of subject and object relative clauses. The children were found to slow down more at the relative clause/main clause boundary when reading or listening to sentences containing object relatives such as The prince that the king taught rode the car to the palace than those containing subject relatives, a finding that is familiar from the monolingual adult processing literature (e.g., King & Just, 1991). Subject–object relatives are more difficult to process than subject–subject relatives because an object gap is further away from its filler than a subject gap (Gibson, 1998), and because they involve perspective shifts (MacWhinney, 1982). Children with a high digit span processed the critical region more slowly than low digit-span children, which Booth et al. (2000) suggest may be due to the former group's processing complex sentences more effectively by making use of their short-term memory store during the processing of difficult regions. On the assumption that the extra slow down observed at and immediately after the embedded verb in sentences containing object relatives reflects the mental reactivation of the direct object filler at this point, Booth et al.'s results, like Roberts et al.'s (2006), suggest that differences in short-term memory may affect children's ability to link a filler to its associated gap in real time.

In sum, the above findings demonstrate that children do not differ from adults in their ability to establish syntactic dependencies during on-line sentence comprehension, but that both children's and adults' processing of such dependencies may be influenced by their working memory capacity.

L2 processing of filler-gap dependencies

Previous studies on L2 learners' knowledge of wh- movement and subjacency have not produced consistent results (see Chapter 7 of Hawkins, 2001, for an overview), giving rise to the hypothesis that processing problems may be responsible for L2 learners' difficulties with certain types of extraction (compare Juffs & Harrington, 1995, 1996). Results from earlier studies investigating the processing of filler-gap dependencies in the L2 provide evidence for on-line filler integration but are ambiguous with respect to the question of whether L2 learners apply a verb-driven or structure-based gap-filling strategy (Juffs, 2005; Juffs & Harrington, 1995; Williams et al., 2001).

To dissociate verb-based integration effects from syntactic gap filling, Marinis, Roberts, Felser, and Clahsen (2005) carried out a self-paced reading study investigating how L2 learners of English from different language backgrounds process sentences involving long-distance wh- dependencies of the kind shown in Examples 3a and 3b. Two corresponding nonextraction sentences served as control conditions.

For sentences such as Example 3a that involve wh- extraction from a complement clause, an intermediate gap is assumed to be present at the intervening clause boundary breaking the long dependency up into two shorter ones (Chomsky, 1977). No such intermediate gap is present, on the other hand, in sentences such as Example 3b that involve extraction across a complex NP. Although the linear distance between filler and subcategorizer is the same in both Examples 3a and 3b, integrating the filler with its subcategorizing verb should be facilitated by the availability of an intermediate gap at the clause boundary, if the processor consults a mental representation of the filler at this point during processing. Results from an earlier study by Gibson and Warren (2004) using similar materials indicate that adult native speakers of English do indeed make use of such intermediate gaps during sentence comprehension.

Marinis et al.'s (2005) experimental sentences were presented visually in a segment by segment fashion, and were followed by a comprehension question. Participants who postulate intermediate gaps should find it easier to integrate the filler with its subcategorizer in sentences like Example 3a than in sentences like Example 3b, which should be reflected in shorter reading times on the segment containing the subcategorizing verb (=segment 5) in Example 3a than in Example 3b. An overview of Marinis et al.'s results is given in Table 6.

All participant groups took longer to read segment 5 in the two extraction conditions than in the nonextraction conditions, which reflects the extra processing cost associated with integrating a filler with its subcategorizer at this point. Only the native speakers showed a significant interaction between extraction and phrase type on this segment, however, indicating that the presence of an intermediate gap facilitated filler integration for this group. That is, the native speakers' reading times at the segment containing the subcategorizing verb were shorter for sentences that contained an intermediate gap than for those that did not, whereas no such difference was observed between the two nonextraction conditions. The native speakers also took longer to read the segment containing the complementizer that in Example 3a than in the corresponding nonextraction condition, which is consistent with the hypothesis that the filler is mentally reactivated at this point during parsing.

Although Marinis et al.'s (2005) results from the native speakers replicate Gibson and Warren's (2004) findings, neither the learners from wh- in situ backgrounds (L1 Chinese and L1 Japanese) nor those from wh- movement backgrounds (L1 German and L1 Greek) showed any intermediate gap effect. In other words, there is no evidence that the learners mentally reactivated the filler prior to the processing of the subcategorizing verb in either of the two extraction conditions. This suggests that contrary to the native speakers, the learners did not observe subjacency during processing but tried to establish a direct link between the filler and its subcategorizer in both extraction conditions instead, even if the subjacency principle is operative in their L1. Note that the learners' verb-driven processing strategy did not seem to affect their ability to comprehend the experimental sentences, though, as witnessed by the fact that they were as accurate as the native speakers in answering the end of trial comprehension questions.

Results from previous studies are consistent with the assumption that nonnative comprehenders employ a lexically driven strategy when processing syntactic dependencies. In two on-line grammaticality judgement experiments, Juffs and Harrington (1995) examined how Chinese-speaking learners of English processed subject and object extractions. The learners showed difficulty with grammatical sentences involving subject extraction such as Who did Anne say________likes her friend? but not with object extractions (see also White & Juffs, 1998). The authors suggest that the learners' selective difficulty with subject extractions may be due to reanalysis problems. In sentences involving subject extraction, the filler is likely to be mistaken initially for the object of the matrix verb, an analysis that must subsequently be revised. Whereas this kind of reanalysis does not seem to cause much processing difficulty for native speakers, it may well do so for nonnative comprehenders (compare Juffs & Harrington, 1996). An explanation in terms of reanalysis difficulty has been called into question, however, by Juffs (2005), who carried out a reading-time study using similar materials to those of Juffs and Harrington (1995) with Chinese-, Japanese-, and Spanish-speaking learners of English. Here the learners showed more difficulty with grammatical subject extractions from finite than from nonfinite clauses, with the Japanese group showing greater processing difficulty than the Chinese- or Spanish-speaking learners. Differences in the learners' working memory span, however, did not appear to influence the learners' processing behavior. According to Juffs (2005), a possible alternative explanation for the learners' problems with wh- extraction from finite clauses might be that they were confused by the co-occurrence of two finite verbs.

In another reading-time study, Williams et al. (2001) examined Chinese-, Korean-, and German-speaking learners' of English processing of sentences that involved adjunct extractions such as Which friend did the gangster hide the car for________late last night? Plausibility was manipulated such that the fronted wh- phrase was either a plausible or an implausible object of the verb hide. All participant groups showed elevated reading times at the postverbal noun in the plausible condition compared to the postverbal noun in the implausible condition. These results indicate that the learners, like the native speaker controls, initially analyzed the wh- filler as the direct object of the verb, and that the subsequent reanalysis was more difficult if the wh- filler was a plausible direct object. The observed “filled-gap” effect suggests that both native speakers and L2 learners attempt to integrate a filler with a potential subcategorizer at the earliest opportunity.

Taken together, the above studies provide evidence that L2 learners are like native speakers in that they attempt to integrate a displaced constituent with its (potential) subcategorizer when this is encountered, and in that they show sensitivity to plausibility information when processing sentences containing filler-gap dependencies. Although the results from Juffs and Harrington (1995), Juffs (2005), and Williams et al. (2001) do not allow us to distinguish between verb-driven and structure-based gap-filling strategies, the absence of any intermediate gap effect in the learners in Marinis et al.'s (2005) study indicates that L2 learners establish long filler-gap dependencies using direct lexical association rather than structure-based gap-filling, regardless of their L1 background.

Preliminary summary

Comparing how children and adult L2 learners process syntactic dependencies has revealed differences between the two types of language learners. Child L1 learners have been found to apply essentially the same parsing routines as mature native speakers when processing syntactic dependencies, and to reactivate displaced constituents at gap positions that are nonadjacent to their lexical subcategorizer. Moreover, there is evidence that children's ability to establish filler-gap dependencies in real time is affected by their short-term memory capacity. Although L2 learners seem to be able to link a fronted constituent to its subcategorizer during processing, there is no evidence that the dependencies thus established are mediated by syntactically defined gaps. Instead, the results from the small number of published studies on processing filler-gap dependencies in the L2 are consistent with the hypothesis that L2 learners establish such dependencies primarily on the basis of lexical–semantic and pragmatic information.