Chapter 1 Introduction
1.4. Morphosyntactic processing in bilingual adults
Research in language processing in adult bilinguals has been shaped by the question as to whether sentence processing in the L1 and L2 is qualitatively different or not. The distinction has been framed between monolingual speakers of a language and late sequential bilinguals, often with predominantly classroom exposure, in other words non-native speakers, or L2 learners. One of the most influential theoretical proposals is the Shallow Structure Hypothesis under which L2 learners differ from natives in that their processing is less syntactic in nature as the grammatical constraints are less robust than in monolingual natives (Clahsen & Felser 2006a; 2006b; 2006c). Although L2 learners’ syntactic representations were assumed to be shallower and less detailed than those of native speakers (Clahsen & Felser 2006a: 3) and that this entailed fundamental and qualitative differences between the two groups (Clahsen & Felser 2006b: 118), this does not entail deficit. Clahsen & Felser claimed than L2 learners underuse syntactic information (2006a: 21) and are thus, more reliant on information from other sources (e.g. lexicon, pragmatics, intonation) which is assumed to be inadequate compensation. This account originated from the long-standing observation that L2 learners
34
have difficulty with grammar and is grounded in research into real time processing in L2 learners. The latter has showed nativelike attainment for some linguistic features, including some aspects of morphosyntax but also persistent difficulties with other aspects of morphosyntax or complex structures even in learners with a high degree of proficiency (see Clahsen & Felser, 2006a for a review). The Shallow Structure Hypothesis was an attempt to explain differences observed in L2 processing which could not be accounted for under the assumption of a global deficit in L2 learners’ capacities but also to integrate L2 processing within existing psycholinguistic approaches to acquisition and processing (Clahsen & Felser, 2018).
One study which lends support to the Shallow Structure Hypothesis is by Marinis, Roberts, Felser & Clahsen (2005) which investigated processing of wh-questions with intermediate gaps as in (15a) in comparison to sentences without the intermediate gap as in (15b) in English L2 learners.
(15a) The nurse whoi the doctor argued ei ' that the rude patient had angered ei is refusing
to work late.
(15b) The nurse whoi the doctor’s argument about the rude patient had angered ei is
refusing to work late.
The difference between the two sentences is that, in (15a), there is a possible filler site for who after the verb “argued”; it is likely that the parser will utilise subcategorization information about the verb “argued” and interpret the filler as the subject to an upcoming verb, but would then need to revise the interpretation and insert the filler in the object position of the verb “angered”. This earlier occurring possibility is not possible in (15b). Marinis et al used a self-paced reading task followed by a comprehension question similar to Gibson & Warren (2004) but with L2 learners whose native language was Chinese, Japanese,
35
German or Greek with at least an upper intermediate level of proficiency. Native English speakers were also tested as controls. At the segment of the intermediate gap, Marinis et al (2005) found a slowdown in reaction times for the native speakers for sentences with an intermediate gap in comparison to the respective segment in the control sentences. Conversely, they found a facilitation effect, i.e. faster reaction times, two segments thereafter, at the subcategorising verb “angered” for sentences with an intermediate gap compared to sentences without one. This was not found for any of the groups of L2 learners. This was interpreted as evidence that the L2 learners, irrespective of the L1, did not posit any intermediate gaps during processing as expected by filler-driven structural accounts suggesting they “underuse syntactic information” (2005:70). However, Dedykspotter, Schwartz & Sprouse (2006) reanalysed some of the data from this study and found the same effect for the L2 learners but at the subsequent segment indicating the effects were delayed but present. This, however, may reflect the storage of the filler in working memory. A limitation of the Marinis et al. (2005) study is that two types of sentences differed in the word length preceding the critical segment due to differences in structural complexity.
Differences between L2 learners with naturalistic relative to those with classroom exposure were found in another study by Pliatsikas & Marinis (2013a). In this follow up study to Marinis et al. (2005), Pliatsikas and Marinis compared L2 learners and tested wh-questions with intermediate gaps as in “The politician who the journalist predicted that the government report would bother is calling a press conference” using a self-paced reading task and the same material as in Marinis et al. (2005). At the intermediate gap segment, only the native speakers showed a slowdown for sentences where there was an extraction suggesting the L2 learners did not process the intermediate gap. All three groups showed a slowdown for the sentences with an extraction relative to those without one at the following segment. At the segment of the subcategorising verb, the native speakers and the L2 learners who had
36
naturalistic exposure to English showed facilitation, i.e. faster reading times, for the sentences with a VP extraction relative to those with an NP extraction. This was absent for the L2 learners with only classroom exposure and is suggestive of a more lexically-driven processing. The facilitating effect of the intermediate gap was interpreted as evidence for the use of syntactic cues. The results from Pliatsikas & Marinis indicate that the nature of processing in L2 learners changes with exposure; learners who become bilingual with a more naturalistic exposure become more nativelike in how they process sentences even for structures such as intermediate gaps which are infrequent in spoken English. The results challenge the Shallow Structure Hypothesis as the predictions it makes are empirically confirmed only for a subset of L2 learners relative to its predictions; only the advanced L2 with classroom exposure showed different and less structural patterns of processing to the natives8.
In another study, Felser, Roberts, Marinis & Gross (2003) investigated how L2 learners with German or Greek as their native language process sentences in English with ambiguous NP attachment such as “The dean liked the secretary of the professor who was reading a letter”. The ambiguity in this sentence is who was reading the letter, the secretary or the professor and speakers of different L1s tend to have a preference for either one or the other depending on the language. Both groups undertook an offline task where they heard sentences as in the example and had to answer comprehension questions relating to the attachment and then did a self-paced reading task with the same type of sentences. For both tasks, Felser et al. manipulated the preposition to be either “of” or “with”. Moreover, in the self-paced reading task, the number of the NPs was manipulated so that only one type of attachment was possible and this became apparent at the auxiliary verb in the embedded clause. In other
8For different results on processing of inflected verb forms where both L2 learners with
naturalistic exposure and ones with classroom exposure perform in a nativelike manner, see Pliatsikas & Marinis (2013c)
37
words, if the participants’ preferred attachment was inconsistent with the one available at the point of the auxiliary, due to subject verb agreement, a slowdown is anticipated. The L2 learners whose native language was German showed a preference to attach the relative clause to the second NP with the preposition “with”. This was evidenced by decreased reaction times at the critical segment when subject-verb agreement forced NP2 attachment. This was also found for the Greek speakers. Native speakers also showed this preference for sentences but with both the preposition “with” and “of”. The L2 learners, however, showed no differences between conditions for “of” suggesting there was no particular attachment preference for sentences with the genitive. Similar results were found for the off-line comprehension questions. In sum, Felser et al. argued that the participants could use lexical information from the preposition “with” but this was not possible with “of”. Felser et al. interpreted the results to show that L2 learners could use lexical or semantic information during processing. The difference between the two prepositions is assumed to be that “with” provides semantic information as it can assign a thematic role whereas “of” does not. Furthermore, they argue that there was little evidence for an influence from the L1 as German and Greek speakers displayed the same processing patterns even though the two languages have contrasting attachment preferences. The native NP2 attachment preference in the “of” conditions was interpreted as the result of attaching the NP to the structurally closest constituent to the predicate phrase (here, the NP2 interpretation). In this sense, the natives followed a syntactically-driven strategy of interpreting ambiguity but the L2 learners did not. Therefore, Felser et al (2003) interpreted the findings as in moderately support in shallow structural processing in L2 learners9. Papadopoulou and Clahsen (2003) conducted a similar
9It should be noted that not all studies which examine processing of relative clauses support
this; for example, Hopp (2014) shows that relative clause attachment is dependent on working memory capacities and lexical processing with L2 learners with higher performance on both measures, on a par with native speakers, showing nativelike patterns of processing.
38
study to Marinis et al. but in Greek testing L1 and L2 Greek (L1 Spanish, German or Russian) and found similar results.
An alternative theoretical account which has been applied to processing to account for differences between adult natives and L2 learners is the Interface Hypothesis10 (Sorace, 2011). Under the latter, L2 learners may not reach nativelike attainment for linguistic phenomena which involve a combination of syntactic knowledge and other cognitive abilities. As applied to processing, this entails an inability for the L2 learners to integrate syntactic information with other information in real time. Several studies have been argued to support the Interface Hypothesis. For example, Belletti, Bennati & Sorace (2007) studied the production and interpretation of overt, null and post-verbal subjects in Italian in L1 English L2 Italian adults. All 3 are grammatical in Italian but the use of each one is argued to be determined by pragmatic, in particular discourse factors. In a series of elicited production tasks, Belletti et al. demonstrated that adults with near native L2 Italian used overt pronominal subjects more frequently than L1 native Italians and were more likely to interpret overt subject pronouns in an embedded clause as coreferential with the subject of the matrix clause relative to the control group. On the other hand the L2 learners produced more post- verbal subjects than the native controls but just as many null subjects, even though the latter aren’t available in English. In sum, Belletti et al. (2007) show that even at high levels of L2 proficiency, some difficulties persist for features of morphosyntax the use of which interacts with discourse. In other words, Belletti et al. argue that the morphosyntactic features have
10Both the Shallow Structure Hypothesis and the Interface hypothesis account for differences
between native speakers and L2 learners of a language in terms of the information utilised during processing but do not assume differences in representation. Other models do posit such differences but are not discussed further as they are less relevant to this thesis - for example, the declarative/procedural model distinction (Ullman 2001, 2005, 2015). However, support for this type of model is derived from studies on morphological processing rather than morphosyntactic processing (e.g. Pliatsikas & Marinis, 2013a, 2013c, possibly 2013b as well, albeit less directly)
39
been acquired for L2 Italian; within a principles and parameters approach, they argue that the null-subject parameter has been set for Italian in the L2 learners. The source of the difficulty is assumed to be the relevant settings of the L1 which are assumed to be accessed in real time processing of the L2.
In another study supporting the Interface Hypothesis in adult L2 learners, Sorace & Filiaci (2006) examined the interpretation of anaphora within bi-clausal sentences in a similar population of L2 Italian learners. Sorace & Filiaci manipulated the subject pronoun to be either null or overt and which clause was the first in the sentence (i.e. the main clause preceded the subordinate clause and there was forward anaphora or the subordinate clause preceded the main clause and there was backward anaphora). The study used a picture verification task as in Belletti et al. where participants needed to select a picture to match the sentence they heard. Sorace & Filiaci found no differences between native speakers of Italian and respective L2 learners in terms of null subject pronoun interpretation; for forward anaphora both groups interpreted the null subject in the embedded clause equally as coreferential to the subject or complement of the main clause but mainly as coreferential to the subject of the main clause in the backwards anaphora condition. Differences between the two groups emerged in the interpretation of the overt subject pronouns. In the forward anaphora condition, the native speakers were more likely to interpret the overt pronoun as coreferential with the complement of the main clause whereas the L2 learners preferred to interpret it as coreferential with the subject of the main clause. In the backwards anaphora condition, the L2 learners interpreted the pronoun as the subject of the main clause, something the natives rarely did, but the natives interpreted it as referring to an extrasentential antecedent, which was also uncommon for the L2 learners. Sorace & Filiaci argue that advanced L2 learners had a representation of the null and overt subjects in Italian
40
but had difficulties using the pragmatic constraints that differentiate between the two in real time.
Other accounts advocate no qualitative differences between processing in the L1 and the L2 but posit that L2 processing is cognitively more demanding (e.g. McDonald, 2006; Hopp 2006, 2010). Such accounts consequently predict that differences between L1 and L2 processing can be predicted by individual differences, i.e. they will be larger in L2 learners with more limited memory capacity, or may become more evident in tasks depending on the cognitive demands involved. In a study involving a range of tasks, McDonald (2006) compared L2 learners of various linguistic backgrounds with native speakers of English. The tasks involved a working memory span, a gated word recognition task (in which the segments of the word were progressively presented), a word detection task to measure speed of processing and a grammaticality judgement task, all of which were administered in English. With the exception of response speed on grammaticality judgement, the L2 learners underperformed on all measures relative to the native speakers. McDonald carried out a second experiment with native speakers. In this experiment, the participants were placed under stress conditions by manipulating either the working memory load (high vs. low; participants needed to remember four or seven digits) or speed (normal speed vs. compressed speech) or making decoding harder (by listening through white noise) with each participant being assigned to one condition. Participants undertook the same tasks as in the first experiment. In all measures except the low memory span condition, native speakers’ performance dropped and for the noise condition performance on the grammaticality judgement task was not significantly better than that of the L2 learners from the first experiment that were under no condition of “stress”. This indicates that L2 processing is more resource intensive that native language processing.
41
More indirect evidence in support of this notion comes from studies testing L2 learners which show more nativelike patterns of processing with near nativelike proficiency. One study in this direction comes from Hopp (2006); in this study, advanced L1 English and Dutch speakers of German were compared to native speakers using a self-paced reading task to test subject-object ambiguities. In the self-paced reading task, the ambiguous sentences where those which were not disambiguated by case but by subject-verb agreement which was clause final as the clauses were embedded. All groups showed lower accuracy for those sentences which were ambiguous but differed in the reading times; only L2 learners with a near native proficiency score in German as established by a close test behaved like the natives and showed a slowdown at the point of disambiguation.
In another study, Hopp (2010) compared English-German, Spanish-German and Russian- German adult speakers to native German speakers using a range of tasks on subject-verb agreement, grammatical gender agreement and case marking. In a grammaticality judgement task, near-native L2 learners differed from advanced L2 learners and clustered with native speakers in that they did not accept sentences with an ungrammatical word order. Furthermore, in a self-paced reading task similar to the one in Hopp (2006), near-native L2 learners behaved similarly to native speakers in that they were sensitive to both case marking and subject-verb agreement (similarly to Hopp, 2006), as they showed a slowdown at critical segments were there was disambiguation. This was not found to be the case for the advanced L2 learners of any of the language combinations11.
The studies reviewed so far have treated the differences in L2 proficiency as categorical. More recent studies have examined proficiency and various measures of L2 competence and cognition as a continuum in order to predict individual differences. One such study is Hopp (2014) who examined relative clause attachment in German-English speakers compared to
42
English native speakers on an eye-tracking reading study of temporarily ambiguous sentences. Hopp showed that attachment preference was modulated by reading span and word level processing (as evidenced by a lexical decision task). Non-natives matched with natives in capacity for working memory and lexical processing demonstrated nativelike parsing. Reading times were significantly higher at the disambiguating region and overall for sentences where the ultimate interpretation favours high attachment than low attachment. As lexical recognition automaticity increases, preferences become more nativelike, i.e. favouring low attachment. These findings are directly contradictory to the Shallow Structure hypothesis as near-native L2 learners show parsing that is similar to that of natives. Moreover, the impact of reading span and lexical decision speed suggests that processing is modulated by factors of language ability. This is again not consistent with the Shallow Structure Hypothesis as the differences between natives and non-natives are shown to be potentially overcome even in cases of late or non-naturalistic exposure. Similar results have been found using an eye-tracking reading study with the same population of L2 learners which tested the processing of subject-object garden path sentences in English (Hopp, 2015).