Abstract
We investigated whether 68 non-native, tutored beginning L2 learners of Italian – with alphabetical and non-alphabetical L1s – discriminated between sentences containing target-like and non-target-like auxiliaries. We questioned whether learners’ choices could be informed by a grammatical rule, frequency of auxiliaries in the input or whether both grammatical and statistical knowledge could be eclipsed by processing difficulties. Eye-tracking and timed acceptability judgment data showed that – unlike native speakers – these L2 learners were unskilled readers of the target language and that their processing was still non-optimal. In particular, they did not process “core” (i. e., strongly agentive and inherently telic) and “peripheral” (i. e., less semantically specified) intransitive predicates differently, nor did they do so with “matching” and “mismatching” predicates. Frequency and transition probabilities speeded up learners’ decisions on acceptability, but did not affect response accuracy or reading patterns. Finally, recency and length of classroom instruction – unlike learners’ L1, duration of stay in Italy, and proficiency level – positively correlated with greater nativelikeness in the processing of auxiliaries. Our results indicate that beginning L2-Italian learners – as long as their processing is still non-optimal – are not sensitive to the consequences of the unaccusative/unergative split at the syntax-semantics interface.
1 Introduction: Auxiliary selection in Italian
Auxiliary selection (AuxS) between avere ‘have’ (A) and essere ‘be’ (E) in compound past tenses of intransitive verbs is a diagnostic for the unaccusative/unergative split in Italian.[1] In modern standard Italian,[2] most verbs displaying auxiliary E (like arrivare ‘arrive’ in sentence (1)) are unaccusative, while most intransitives displaying auxiliary A (like telefonare ‘phone’ in sentence (2)) are unergative:
| Elena | è | arrivata |
| Elena | isaux3sg | arrivedpastp |
| ‘Elena arrived/has arrived’ | ||
| Elena | ha | telefonato |
| Elena | hasaux3sg | phonedpastp |
| ‘Elena phoned/has phoned’ | ||
When the auxiliary is E, the past participle (the part following the auxiliary) agrees with the grammatical subject. Therefore, arrivata ‘arrived’ is used when the subject is singular feminine, arrivato when the subject is singular masculine, and arrivati and arrivate when – respectively – the subjects are plural masculine and feminine. In contrast, when auxiliary is A, the past participle is invariable.[3]
In this study, we question whether non-native (NN), tutored beginners and low-intermediate learners of Italian may have different representations for A and E. The method adopted combines eye-tracking measures and acceptability judgments. This combination is intended to test whether learners’ online decisions about the acceptability of sentences containing target-like and non-target-like auxiliaries are driven by a grammatical rule, by frequency or by none of these factors because learners’ processing is still non-optimal. The novelty of this study is twofold. First, we separately factorized processing and representational difficulties. Auxiliaries may be difficult to learn not because L2 learners cannot represent a grammatical rule or because learners are not sensitive to the frequency of auxiliaries in the input, but because they are unable to process the link between the auxiliary and the past participle in Italian. Second, we assumed that learners’ knowledge of auxiliaries can be either grammatical (based on learners’ knowledge of unaccusative predicates) or statistical (based on frequency and transition probabilities between auxiliaries and past participles).
The article is structured as follows. Section 2 defines “non-optimal processing” in the context of Second Language Acquisition and AuxS. Section 3 outlines the differences between an L2 auxiliary knowledge driven by optimal processing of a grammatical rule and an L2 knowledge driven by optimal processing of auxiliary frequency in the input. In Section 4, the research question and the predictions are outlined. In Section 5, the experiment is described. Finally, in Sections 6 and 7, the results are presented and discussed.
2 Non-optimal L2 processing of auxiliaries
Auxiliaries may be difficult to learn not only because L2 learners cannot represent a grammatical rule or because they are not sensitive to their distribution in the input, but also because learners are still unable to process the link between the auxiliary, the past participle and all the other elements of the sentence that are relevant for AuxS. L2 beginners – especially those from non-alphabetical L1s – are unskilled readers of the target language. Their reading patterns often reflect processing routines typical of L2 learning processes, and this might be partly independent of developing L2 grammars or frequency-related factors. It is thus possible that experimental measures – such as eye-tracking – will reflect non-optimal L2 processing of auxiliaries rather than representational or distributional deficits. In a situation characterized by non-optimal processing, a number of phenomena signal the presence of a developmental process which is called “attunement”. The presence of attunement neither necessarily signals a lack of underlying knowledge of the L2 grammar, nor does it automatically reflect learners’ insensitiveness to the distributional features of the target-language input.
Let us define what “non-optimal” means first in the context of L2 processing and then in the context of AuxS. Based on ambiguity resolution studies of relative clause attachment, Frenck-Mestre (2002; 2005) suggests that lower automatization and processing efficiency in L2 reading are responsible for developed habits such as slowness, phonological activation (spelling aloud), re-readings, increased number of regressions, word-by-word fixations (no skipping), and preference for lexical over morphological cues (see also VanPatten 2007: 117). For example, native speakers usually skip short (mono or disyllabic) words, as well as highly predictable and function words, including auxiliaries E and A; almost all their saccades (eye-movements) are forward; they fixate just once; the landing sites of fixations are at the left half of the word, at a regular distance of 7–9 characters, which corresponds to the perceptual span (where vision is more acute and linguistic information is extracted more rapidly and effectively) (Irwin 2004; Staub and Rayner 2007; Shillcock 2007). Conversely, L2 learners often fixate almost every letter of every word, do not selectively and consistently skip upcoming words, re-fixate each word two or three times.
In the context of AuxS, the processing habit described above – especially longer fixations of single words – may drive learners to consider auxiliaries in isolation, and thus may prevent them from projecting the relevant grammatical or statistical features to other elements in the sentence. In fact, in order to select the proper auxiliary, learners should know that all occurrences of an auxiliary such as E joined together with all occurrences of a variable (suffixed) form, such as arrivato ‘arrived’, are instances of the same abstract construction. Beginners – even if they are exposed to the input of the target-language for a long time – might find it difficult to learn that such words form a pair. This is because of their tendency to read and process words in isolation. Another factor of non-optimal processing is the existence of a possible natural advantage for left-to-right alphabetical readers over non-alphabetical readers (e. g., Chinese students). Firstly, it seems that the perceptual span is smaller for Japanese and Chinese readers than for English readers (2–3 characters vs. 7–8 characters, respectively) (Inhoff and Liu 1998). If, for these reasons, the L2 processing of Chinese subjects were to be less optimal than other subjects, they would be less likely to exploit the whole length of the perceptual span to extract information. Given the different graphical status of Chinese words, L1 Chinese subjects might experience difficulties in recognizing the place where they can automatically get different (lexical and morphological) pieces of information from a single word, in the case of our study, the difference between auxiliary A and E.
Finally, non-optimal processing should be considered a temporary condition. As learners’ proficiency increases, also processing strategies may evolve, less efficient reading habits are gradually abandoned and the footprints of L2 developing grammars and of frequency begin to be clearly visible through the analysis of processing measures.
3 The knowledge of auxiliaries in L1 and L2 Italian
As many other features of the target language, L2 learners’ knowledge of AuxS may be driven by a grammatical rule or the statistical distributional features of auxiliaries in the input (e. g., Rosemeyer 2014). This means that L2 learners may select the target-like auxiliaries because they have intuitions about split intransitivity, or because they have been exposed extensively to L2 input and/or also to classroom practice where some auxiliary- past participle combinations are more frequent than others. We define the former “grammatical knowledge” and the latter “statistical knowledge” of auxiliaries.
3.1 Grammatical knowledge of auxiliaries
Native speakers (NSs) of Italian may select auxiliary E or A because they know the difference between unaccusative and unergative verbs. The rule for AuxS may be syntactically or semantically driven (Alexiadou et al. 2004; Bentley 2006). In the former case, the syntactic status of the sole argument matters: the verb is unaccusative if it includes a subject which is base-generated in the object position. For instance, the auxiliaries E and A in sentences (1) and (2) (displayed in Figure 1) signal, respectively, the antecedent-trace relationship – or the lack thereof – between the surface subject and the base-generated object (‘Elena’).
In the semantic account, the meaning of verbs decides which class an intransitive verb belongs to. For instance, Italian directed motion verbs (such as arrivare ‘arrive,’ uscire ‘exit,’ entrare ‘enter’) select E, while manner of motion verbs (such as camminare ‘walk,’ pattinare ‘skate’) select A.[4] The syntactic and semantic views are compatible. However, Sorace (2000) and Keller and Sorace (2003: 58) noticed that Italian predicates taking both auxiliaries – for example, meteorological verbs such as piovere ‘rain’ or nevicare ‘snow’ (Benincà and Cinque 1992) – pose a problem for the syntactic approach, and that other predicates taking A or E regardless of their meaning challenge the semantic approach.
Finally, the intransitive split has been accounted for as a lexicon-syntax interface phenomenon (Van Hout 2004; Randall 2007). At the interface, the aspectual features, Θ-roles (verb semantics), NPs, and adjuncts – which are encoded by different items at different layers between
The structural representation of sentences (2) at the syntax-semantics interface.
AuxS and the intransitive split in Italian are a close match, to the extent that MacKenzie (2006: 103) observes that “as far as perfect AuxS is concerned, the Ergative Analysis and Unaccusative Hypothesis are inspired primarily by the Italian situation”. AuxS has indeed a privileged status as a diagnostic of unaccusativity in Italian because its effectiveness – unlike ne-cliticization, absolute participle, and Verb-Subject inversion – does not depend on the speaker’s intuitions about the position of the subject in the deep structure (Levin and Rappaport Hovav 1995: 18). Therefore, the A|E alternation could provide positive and systematic evidence of the intransitive split also to L2 learners of Italian (Oshita 2001).
Actually, some authors claimed that NSs and L2 learners are sensitive to the interplay of different factors at the syntax-semantics interface. One of such alternative approaches considers telicity, agentivity, and the notion of “gradient unaccusativity”. The Auxiliary Selection Hierarchy (ASH) (also known as SIH, Split Intransitivity Hierarchy; Sorace 2000; 2004; 2015) labels inherently telic verbs (such as uscire ‘exit,’ arrivare ‘arrive’) and strongly agentive verbs (such as lavorare ‘work,’ telefonare ‘phone’) as core unaccusative and core unergative verbs, respectively. Instead, less semantically specified verbs (e. g., those denoting the existence of a state, such as esistere ‘exist’, durare ‘last’) are called peripheral verbs. All verbs are thus placed along a gradient, depending on the extent to which they encode telicity/agentivity lexically or compositionally, at a phrasal level. For example, verbs like arrivare ‘arrive’ and lavorare ‘work’ are – respectively – inherently unaccusative and unergative regardless of the context. In contrast, the unaccusative/unergative behavior and the consequent choice of either auxiliary of the verb volare ‘fly’ can be function of both the animacy of the grammatical subject (Elena vs. l’aereo ‘the plane’) and telic vs durative sentence completion, as in sentences (3) and (4):
| Elena | è | volata | a | New York |
| Elena | is.aux3sg | flown.pastp | to | New York |
| ‘Elena flew/has flown to New York’ (unaccusative) | ||||
| L’aereo | ha | volato | per | ore |
| The plane | hasaux3sg | flownpastp | for | hourstemp_adv |
| ‘The plane flew/has flown for hours’ (unergative) | ||||
The ASH predicts that the auxiliaries of core verbs have primacy in acquisition (Keller and Sorace 2003: 60–61; Sorace 2004: 268). Sorace (2015) specifies that ASH is, by itself, a generalization and not a theory and that its basic tenets are not in contradiction with the Unaccusative Hypothesis (which is by nature dichotomic and incompatible with the idea of a gradient). Indeed, gradient variance and categorical invariance (predicted by the Unaccusative Hypothesis) are supposed to co-exist in languages and in the human mind in general.
As to the empirical evidence, corpus-based and behavioral data draw a contrasting picture. Ježek and Rastelli (2008) examined 65 different intransitive verbal types corresponding to 470 verb tokens produced in a written description of film scenes by 250 American students of L2 Italian at different levels of proficiency (50 % advanced, 38 % intermediate and 10 % beginning learners). They observed that between 20 % and 50 % of the time, auxiliaries were omitted or mistaken, as an inverse function of learners’ proficiency. Especially auxiliary A use was overextended to E use, probably due to L1 English influence. Behavioral data suggest something different. By using the magnitude estimation technique, Sorace (1993) found that 24 English and 20 French advanced learners of Italian performed like native controls in an acceptability judgment task involving the choice of auxiliaries. Kraš (2010) assessed the linguistic intuitions on auxiliaries of 16 highly proficient adult Croatian learners of L2 Italian by using speeded acceptability judgment tasks. Also in this case, means of acceptability scores and response times showed that subjects distinguished between A and E with respect to the intransitive split. The footprints of a rule for AuxS in NN Italian were investigated using the acceptability judgment technique (magnitude estimation). The learners quoted in those studies – like those of L2 German evaluated by Keller and Sorace (2003) – were found to be sensitive to the phenomenon of gradient unaccusativity.
3.2 Statistical knowledge of auxiliaries
Native speakers of Italian and L2 learners may be sensitive to the distributional features of auxiliaries in the input. They can recognize that the auxiliary and the past participle (PastP) in compound tenses form the pair [Aux+PastP]. Aux+PastP is a meaningful, statistically significant combination of words where the latter (the past participles) are far more numerous in the input than the former (because in Italian there are only two auxiliaries, A and E). The relative type-frequency of the units that form such constructions is unbalanced and statistical representations of Aux+PastP constructions in a NS’s mind are unlikely to originate from forward predictability (the probability that a word is followed by another) but from backward transition probability (BTP) (the probability that a word is preceded by another) (Onnis et al. 2008). In cases where Aux and PastP frequently co-occur, they form a collocation in which the collocational headedness is not on the head, but on the second element of the collocation (Malec 2010: 129) and the direction of headedness is backwards. For instance, in the reference corpus of written and spoken Italian used in this study (Section 5.2), the probability that the auxiliary ha ‘has’ is followed by squillato ‘rung’ is approximately one hundred thousand times lower than the probability that squillato is preceded by ha (= .42).[5] The consequence is that this Aux+PastP is an asymmetric construction whose predictive element follows the auxiliary (in our case, squillato).
Finally, statistical learning of auxiliaries is a function of a learner’s capacity of tracking not only the transition probabilities such as the one described above (the conditional probabilities of ha ‘has’ preceding squillato ‘rung’ given ha + squillato), but also token and type frequency. Token frequency depends on the number of repetitions of the same Aux+PastP construction in the input. Type frequency depends on how many different instances of the same verb enter the Aux+PastP construction in the input.
4 Research question and predictions
The research question in this study concerns whether L2 learners of Italian can discriminate between target-like and non-target-like Italian auxiliary uses and whether – when doing so – they can parse auxiliaries by a grammatical rule (see Section 3.1) or by relying on statistics (see Section 3.2), or if they cannot process auxiliaries because their processing is non-optimal (see Section 2). The predictions are as follows.
If a syntactic or semantic rule for AuxS is at work, we expect peripheral verbs on the ASH gradient (see Section 3.1) to be processed more slowly than core verbs and non-target-like auxiliaries occurring with core verbs to cause more disruption than those occurring with peripherals. This would occur because – according to the predictions of the ASH (see Section 3.1 – L2 learners are credited to have stronger intuitions for core than for peripheral verbs, and violations occurring with the former would especially alter their reading patterns. We also expect verb syntax and semantics to be a factor in generating target-like acceptability judgments, with subjects selecting E for unaccusatives and A for unergatives, and having clearer intuitions with core rather than with peripheral verbs. In contrast, if the processing of auxiliaries in L2 Italian is driven by statistics (see Section 3.2), we predict higher frequency and BTP between Aux and PastP to speed up reading and to correlate with target-like performance in acceptability judgments, whereas weaker BTP between the auxiliary and the PastP will hinder the immediate recognition of Aux+PastP constructions.
Finally, if learners’ processing is non-optimal (see Section 2), we expect the presence of both grammatical and statistical knowledge of auxiliaries to be shadowed by unskilled reading habits due to learners’ lack of experience with the target language. Namely, we expect NN speakers – regardless of the kind of verb and whether its use is target-like or not – to show slower reading and longer path-length, longer and more numerous fixations, and not to skip the auxiliaries, as NSs do. Finally, we also expect length of exposure to the target language, length and recency of instruction, and L1 features (alphabetical vs. non-alphabetical) to play a role – whether decisive or minimal – in either reading patterns or acceptability judgments (or in both).
5 The study
5.1 Participants
The initial participants in this study were 78 L2 Italian learners (mean age 23.4, range 19–25) and 37 native controls (mean age 22.4, range 18–33). All subjects were right-handed, visually paired undergraduate students, and they were paid for their participation in the experiment. Eight subjects were successively discarded because they failed the lexical test (see below), or due to poor eye tracking data (e. g., because the subject gazed outside of the eye tracker range). The remaining 70 NN participants were divided into three groups. Group-A subjects were European students attending the Erasmus Exchange Program; subjects in groups B and C were attending the Marco Polo Program for Chinese students.[6] L1 Chinese students from group B had finished their Italian course about sixteen months prior to the experiment and were no longer receiving any language instruction, whereas students of group C were attending Italian courses at the time of the experiment. Table 1 summarizes the between-groups variability along seven descriptive dimensions.
Dimensions of between-group variability: (i) n. = number of subjects; (ii) L1 = subject first language; (iii) α = alphabetical L1; (iv) exposure to the target language in Italy (in months); (v) instruction = sum of months of previous instruction and in Italy; (vi) recency of instruction (in months) = “+” = currently attending courses for...months; “−” = have no longer been attending courses since...; (vii) degrees of proficiency (1–6) reflect the levels of the Common European Framework of Reference for languages.
| Group | N. | L1 | α | Exposure (m) | Instruction (m) | Recency | Proficiency |
| A | 23 | German, Romanian, Polish, Croatian, English, Spanish | yes | 3.25 | 7.75 | +3 | A2 |
| B | 23 | Chinese | no | 25.2 | 7.8 | −16.8 | A2 |
| C | 24 | Chinese | no | 3 | 5.4 | +2 | A2 |
| Control | 37 | Italian | yes | na | na | na | na |
The presence vs. absence of an auxiliation system similar to Italian was not factorized in the analysis. Indeed, only a subset of group A subjects had L1s with a productive, semi-productive (e. g., frozen) – binary or unary – auxiliation system comparable to Italian. For example, Romanian – like Italian – distinguish between fi ‘be’ and avea ‘have’ but the former has an adjectival meaning for most speakers (Avram 1994). Spanish has only haber ‘have’ as auxiliary in compound tenses (e. g., the pretérito perfecto compuesto), while the verbal vs. adjectival nature of frozen be-auxiliaries in the English present perfect is debated (Levin and Rappaport Hovav 1995; Sorace 2000). Chinese (i. e., the L1 of Group-B and Group-C subjects) and the native language of some of the subjects in group A only feature modal auxiliaries. For example, Polish features the modal auxiliary mieć, corresponding to the English ‘must’, ‘have to’. Croatian has biti ‘be’, and the modal htjeti ‘want’. Chinese also features modal auxiliaries corresponding to the English ‘can’, ‘must’, and ‘should’. German intransitive verbs – like Italian intransitive verbs – have been claimed to be sensitive to the ASH too and to select either sein ‘be’ or haben ‘have’ depending on telicity and agentivity. However, there are some important differences: for example, unlike in Italian, verbs of motion tend to allow sein ‘be’ even in the absence of a directional phrase that telicizes the predicate (Keller and Sorace 2003: 71).
Subjects’ proficiency was established by means of two consecutive CILS level A2 tests (the official Italian proficiency exams administered by the University for Foreigners of Siena). The test was based on five trials (listening and reading comprehension, grammatical cloze test, written composition, and oral interview).[7] All participants passed the A2 test, and the average score was 80.2 (out of 100). All groups received the same kind of explicit instruction about AuxS by three language instructors using the same textbook. Students encountered auxiliaries on many occasions in the classroom input and they all practiced the same structural drills and focus-on-form/s activities (such as recasts or prompt interventions) on auxiliaries for one month preceding the experiment. All subjects were given the opportunity to familiarize with the words used in the experimental sentences. They were given a list of full words (nouns, adjectives, and verbs lemma) translated into their native language two weeks before the experiment and they were asked to learn them by heart. Subjects’ knowledge of the words was tested through two tests (oral and written) half an hour before the experiment.
5.2 Frequency scores
We used token-frequency of Aux+PastP constructions and BTP scores as measures of statistical learning of auxiliaries. Frequency scores of Aux+PastP constructions were computed as follows. Absolute occurrences were sorted from seven corpora of written,[8] spoken[9] and on-line[10] contemporary Italian. Occurrences were normalized to 1-million size. The representativeness of different genres and media (spoken, written, on-line) was balanced by equalizing token occurrences with the occurrences (tokens) / corpus size (tokens) ratio. After the mean of occurrences and the standard deviation were computed, a normality test with empirical rule[11] was then used to exclude outliers. Finally, the re-calculated means were converted into a ten-point scale. To compute the BTP score, the occurrences of constructions (formed by either auxiliary ha or è and a masculine singular form of PastP) and those of PastPs in the corpora were sorted out separately. Except when occurring in absolute participial constructions, all PastPs were included in the denominator of the ratio. The same procedure with normality test was used to exclude outliers and minimize skewness, and the BTP score was established via the occurrences (Aux+PastP) / occurrences (PastP) ratio. The BTP score increases when auxiliary and PastP are strongly “left-collocated”, but the trend could also be biased by the distribution of PastPs in the input. At least in theory, BTP scores tend to be lower if verbs enter more syntactic classes (because the denominator gets bigger) and/or if the auxiliary fluctuates (because the numerator shrinks).
5.3 Materials
To represent the variability that characterizes AuxS in L1 Italian and the misalignments between syntactic and semantic factors, the experimental design contrasted type (n = 8) and mismatching (n = 8) stimuli (verbs). “Type” stimuli featured verbs for which the unaccusative vs. unergative and the core vs. peripheral distinctions are supposed to be relevant according to the Auxiliary Selection Hierarchy (Section 3.1). “Mismatching” featured verbs that may take both auxiliaries in contemporary written and spoken Italian. In Table 2, all experimental predicates are listed.
Type vs. Mismatching verbs.
| TYPE | MISMATCHING | ||||
| Unaccusative | Unergative | ||||
| Core | Peripheral | Core | Peripheral | ||
| arrivare | rimanere | telefonare | tremare | correre | vivere |
| ‘arrive’ | ‘remain’ | ‘phone’ | ‘tremble’ | ‘run’ | ‘live’ |
| saltare | inciampare | ||||
| giungere | emergere | mentire | cedere | ‘jump’ | ‘stumble’ |
| ‘arrive’ | ‘emerge’ | ‘lie’ | ‘yield’ | continuare | squillare |
| ‘continue’ | ‘ring’ | ||||
| funzionare | mancare | ||||
| ‘function’ | ‘miss’ | ||||
To validate the choice of mismatching verbs in our study, we computed their raw frequency and BTP in L1 Italian written and spoken corpora listed in Footnotes 8, 9, and 10. Table 3 sums up the normalized frequency values of mismatching predicates in those corpora.
Normalized frequency values (per million) and BTP of Mismatching Verbs in spoken and written contemporary L1 Italian corpora.
| Verbs | Normalized frequency A | Normalized frequency E | Backward Transition Probabilities A | Backward Transition Probabilties E |
| correre | 7.84 | 1.50 | 0.52 | 0.48 |
| saltare | 1.1 | 5.7 | 0.09 | 0.91 |
| continuare | 13.62 | 2.43 | 0.82 | 0.18 |
| funzionare | 4.1 | 1.9 | 0.98 | 0.02 |
| vivere | 19.5 | 5.1 | 0.54 | 0.46 |
| inciampare | 0.18 | 0.44 | 0.20 | 0.80 |
| squillare | 0.45 | 0.57 | 0.60 | 0.40 |
| mancare | 1.54 | 7.56 | 0.24 | 0.76 |
One may notice that both auxiliaries A and E of all mismatching verbs are well represented in the input. Therefore, the factorization of such verbs in the experiment is fully justified. Moreover, frequency and BTP showed a mid- to weak strength correlation (Pearson = 0.47 for auxiliary A and Pearson = 0.37 for E). This latter information means that verbs taking either A or E and having similar frequency in the input may not have similar BTPs (this is especially the case of correre ‘run’ and vivere ‘live’). This in turn justifies separating row frequency and BTP as factors for statistical learning of auxiliaries.
The design of the experiment contrasted type vs. mismatching verbs. Type verbs alone also entered a 2×2×2 design and were manipulated for unaccusative vs. unergative, core vs. peripheral, A vs. E conditions. In contrast, mismatching verbs were manipulated separately only for A vs. E condition. In assigning type verbs to either category, we followed Sorace (2000; 2004) (barring giungere ‘arrive, reach’ which is not quoted in those studies). As to the core/peripheral opposition, the analysis of the Italian corpora confirmed that both unaccusative and unergative core verbs display just the expected auxiliary (E for arrivare and giungere, A for mentire and telefonare). The same cannot be said for peripheral verbs, for which fluctuations were negligible (in our corpora, ha rimasto occurs 3 times while è rimasto 23.000 times) or sometimes not attested at all in our corpora (e. g., ha emerso never occurred in the corpora). This latter information confirms that “peripheral” and “mismatching” must be factorized separately.
5.4 Method
In the eye-tracking experiment, each subject had to read 32 experimental sentences (two for each verb, one with auxiliary A and one with auxiliary E) and 32 fillers. Fillers featured both grammatical and ungrammatical sentences (in the latter, ill-formedness related to agreement within the NP or to clitic misplacement). The stimuli comprised three “regions of interest” (henceforth ROIs): onset, critical and spillover zone. The onset zone is the beginning of a sentence, which usually does not contain the target feature of an experiment. The critical zone represents the focus of the research and in the case of current research it comprises both the auxiliary and the PastP. The spillover zone is its final portion (or “sentence completion”) where usually researchers expect the so-called “wrap-up effect” which consists of a reader’s attempt to integrate the semantic and syntactic information that has been processed up to that point in the sentence. This attempt may result in longer reading time and more numerous regressions. The onset and the critical zones were invariable, while the spillover zone changed through different experimental conditions. The sentence onset contained a singular subject (inanimate il tetto ‘the roof’ or animate, e. g., Mario). The subject was always masculine to avoid PastP agreement (Section 1). The critical zone contained a third person singular V (at the Passato Prossimo compound tense) that comprised the auxiliary and PastP. The spillover zone included the V argument/s and an adjunct (PP or AdvP). Sentence (5) is an example of a linguistic stimulus used in the experiment. The complete list of experimental stimuli is provided in Appendix A:
| Luigi/ | ha telefonato/ | a sua mamma con il cellulare |
| onset | critical | spillover |
| Luigi hasaux phoned to his mother with the cell phone | ||
| ‘Luigi called his mother with his cell phone’ | ||
Sentence spillover zones did not remain identical throughout the experimental conditions, except for the E vs. A condition. The pros and cons of this choice were balanced. There were two advantages: (1) based on NSs’ judgments, varying spillover zones (i. e., completions) made the sentences sound much more natural[12] than if the same completions had been used; (b) at least in theory, varied spillovers are less likely to be skipped and this possibly preserves semantic integration or late effect (referred to as “wrap-up effect”). A possible weakness of the choice is that the stimuli of each verb-set did not differ only with respect to the variable-level combinations under investigation. Thus, we could be less in control of which factors affect learners’ choice of either auxiliary.
Sentences were pseudorandomized to avoid undesired adjacency-effects. In order to reduce the risk of learners becoming aware of manipulations and adopting task-specific strategies, every subject only received one item out of a token-set (i. e., single observation for each token-set). Frequency and BTP effect on the length of PastPs were controlled: ANOVAs showed no significant interaction between the number of characters that critical zones are made of (range 7–13) and frequency of constructions (F = 0.2348,
5.5 Procedure
Before starting the experiment, the subjects expressed their written consent; the purpose of the study was not disclosed to them. The experimental task was explained in Chinese to Chinese learners and in English to all the other learners. The wording of the task was as follows: “You will read 64 sentences, individually. Press the right or left keys with your middle finger as soon as you realize that the sentence is grammatical or non-grammatical.” The term “grammaticality” was made explicit as the condition of a sentence being formally flawless, that is, natural and devoid of grammatical errors. There were a warm-up session (10 sentences) and a trial session (10 sentences) to get subjects familiar with the time-course and the keyboard. In this phase, subjects were trained to discriminate among intuitions about ill-formedness and intuitions about complexity. To this purpose, prior to the experiment, the researcher showed each participant examples of long sentences that were natural and acceptable and shorter sentences that were not. This training was aimed to generate grammaticality (‘yes’ vs. ‘no’) judgments based on the criteria of acceptability and naturalness as much as possible. As an eye tracker, we used the Tobii 1750, which integrates all the necessary components (camera, infrared lighting, etc.) into a
5.6 Reading measures
We adopted eight measures for regions (and no measure for single words): (i) number of fixations: total number of fixations detected on the whole sentence; (ii) total reading time (TRT): sum of the durations of all fixations detected on the whole sentence; (iii) total reading times on critical zone (TRT-C): sum of the durations of all fixations detected on the critical zone; (iv) total reading time on spillover zone (TRT-S): sum of the durations of all fixations detected on the spillover zone; (v) first-pass duration on critical zone (FP-C): sum of the durations of all fixations detected on the critical zone during the first pass (i. e., the first sequence of consecutive fixations on the critical zone); (vi) number of fixations on critical zone: total number of fixations detected on the critical zone; (vii) number of regressions: number of backward jumps to any section after the first visit (i. e., after exiting it); (viii) path length (PL): sum of the lengths (in pixels) of the segments connecting each fixation (actually, its position) with the next one. We considered as a fixation a sequence of consecutive gaze samplings by the eye tracker detected within a circle with a radius of 30 pixels, for a minimum duration of 100 ms.
According to most current theories on L1 and L2 processing, we assumed that first passages on a word or a region (early measures) are affected by either lexical access or by gross syntactic anomalies (such as number or gender agreement). If a word is infrequent, unexpectedly collocated, difficult to identify and/or to assign to a phrasal constituent, first-passage fixations are longer, saccades are shorter and regressions to that word/region are more numerous. Instead, when a word is highly predictable, fixation and gaze duration on that word/region are expected to decrease (Rayner and Sereno 1994: 59; Clifton et al. 2007: 344; Shillcock 2007: 97). On the other hand, we can also assume that problems in sentence comprehension and the process of semantic integration have delayed effects on the eye movement record (Pickering et al. 2004).
With respect to the early/late effect divide and to its correlation with syntax, semantics, or statistical representations, L2 processing could be a world apart. Generally, L2 learners might not experience difficulties immediately, but only at the end of the sentence, regardless of the nature of the disruptions. If this is true, late measures – such as total reading time – might be affected more than early measures (first passages) in general, that is, independently of the fact that the anomaly concerns syntax, semantics, or word frequency (Frenck-Mestre 2005). We will thus assume that our reading measures (early vs. late effects) and learners’ sensitiveness (syntactic, semantic, or statistical) are not correlational in nature and must be checked independently. As we have seen in Section 2, another important factor in our study is the existence of a possible natural advantage for left-to-right alphabetical readers (group A) over non-alphabetical readers (Chinese students of groups B and C).
6 Results
6.1 Grammatical knowledge of auxiliaries
As reading measures for testing syntactic and semantic processing, we chose total reading times (TRT), total reading times on critical zones (TRT-C) and first-pass reading times on critical zones (FP-C) because such measures are supposed to be sensitive to gross syntactic and semantic anomalies. Native controls and Group-A subjects (alphabetical L1 readers) processed unergative verbs more slowly than unaccusative verbs (see Bard et al. 2010 for similar results with N speakers), while Group-B and Group-C subjects did not. A multifactorial ANOVA and Tukey post hoc[13] test showed that this difference was significant only in first-pass over critical zone (FP-C) and only for group A (F = 4.21, p < .02), but not for N controls (p = .87). Table 4 shows that all NN subjects processed peripheral verbs more slowly than core verbs, while N controls had slightly faster TRT-C and FP-C.
Average total reading times (TRT), total reading times on critical zones (TRT-C) and first-pass reading times on critical zones (FP-C) × core vs. peripheral conditions and × group (in ms).
| GROUP | TRT | TRT-C | FP-C | |||
| core | peripheral | core | peripheral | core | peripheral | |
| A | 3558 | 3930 | 1347 | 1507 | 470 | 615 |
| B | 6088 | 6644 | 1992 | 2145 | 620 | 672 |
| C | 4154 | 4982 | 1282 | 1394 | 568 | 645 |
| CNTR | 1723 | 1822 | 560 | 528 | 324 | 299 |
Group was a very significant factor (p***) to explain these differences. Tukey post hoc test showed three things: (i) the difference between N controls and L2 learners was significant (p***) in all conditions; (ii) Chinese subjects (i. e., groups B and C) were significantly slower than subjects of group A in all conditions, (iii) Chinese subjects of group B were always significantly slower than Chinese subjects of group C in all conditions, except in first-passes on critical zone. Neither core/peripheral conditions nor their interaction reached significance in all three reading measures (all ps > .20).
For the type/mismatching opposition, only NSs and NN from group A took more time to process mismatching verbs, that is, verbs taking either auxiliary such as inciampare ‘trip up’ or vivere ‘live’. Again, the ANOVA showed that these differences were not significant (all ps > .25), while the group factor was (p***). Namely, the Tukey post hoc tests showed that both the difference between N controls and NN subjects (p***) and the difference between subjects of group A and subjects of groups B and C (p**) were significant.
When auxiliaries were non-target-like, reading times did not increase as expected if reading was disrupted by learners’ sensitiveness to a rule violation. Yet, TRTs significantly decreased with non-target-like auxiliaries in group B (F = 2.86, p < .05), group C (F = 4.05, p < .03) and N controls (F = 2.93, p < .05). To verify whether this decrease was the consequence of faster exits due to the successful spotting of the violations, TRT-S measures (total reading times on the spillover zone) were regressed onto group and testers’ response conditions in a linear regression model (Gries 2015). Both factors taken in isolation were highly significant (p***), while their interaction was not (p = .21). In our data, mean TRT-S were 623 ms longer (not shorter) when subjects answered correctly, even though the post hoc Tukey shows that the variance was significant only for group B (p = .002), but no differences among the remaining groups reached significance (all ps > .25).
Finally, we wanted to check whether the presence of non-target-like auxiliaries increased the probability of regressions to and fixations on critical zones. On average, the presence of non-target-like auxiliaries caused a slight increase in regressions only in N controls and a decrease in all NN groups (whether they were alphabetical L1 readers or not) (see Figure 3). The ANOVA showed that differences between group B and the other groups were significant (p***), but the Tukey post hoc analysis also showed that in no group the decrease was due to the target-like vs. non-target-like auxiliary condition (all ps > .10). Figure 3 shows that NN participants in general regressed significantly more often than N participants on the critical zone (the auxiliary), but only the latter group regressed more often when the auxiliary was non-target-like.
Average n. of regressions on critical zone × kind of sentence.
As to the first passes on critical zones, the difference among all NN and N groups was highly significant (p***), but the target-like vs. non-target-like conditions did not cause significant increases (as shown in group A) or decreases (as shown in group B). Finally, Table 5 and Figure 4 show the percentages of target-like acceptability judgments per group.
Percentage of target-like and non-target-like judgments on auxiliary (type verbs) per group.
Percentage of non-target-like (0) and target-like (1) acceptability judgments per group.
| GROUP | non target-like | target-like |
| A | 35.9 | 64.1 |
| B | 48 | 52 |
| C | 43 | 57 |
| CNTR | 5 | 95 |
A Kruskall-Wallis test for multiple comparisons shows that differences in acceptability judgments were significant between NN subjects and N controls (chi-square = 86.8786, df = 3, p < .001) but they were not significant among NN groups (chi-square = 3.9139, df = 2, p = .14). Figure 5 displays the rate of target-like and non-target-like judgments in the NN and in the N groups per verb category.
Accuracy percentage of acceptability judgments in N and NN groups per verb category.
N controls were more accurate with core verbs than with peripheral, whereas the rate of acceptance of type and mismatching verbs was similar. In contrast, NN participants did not differentiate between core and peripheral verbs. Moreover, only subjects in group A were less accurate with “mismatching” rather than with “type” verbs. To investigate whether such differences were statistically significant and whether unaccusative vs. unergative, core vs. peripheral, and type vs. mismatching conditions were factors for target-like acceptability judgments, we regressed these conditions onto NN subjects’ responses in a linear regression model. Neither the unaccusative vs. unergative contrast (Wald Z = −1.55, p = .12), nor the core vs. peripheral (Wald Z = −0.210, p = .83), nor the type vs. mismatching (Wald Z = 0.56, p = .57) increased the probability of successfully spotting the violations in NN subjects. As a further statistical measure, multiple Chi-square for covariates on acceptability judgments in N controls and in NN subjects confirmed that being a ‘core’ verb was not a predictor (p > .05) of correctness in non-NN subjects, while it was in the N control group (p = 0.0065).
6.2 Statistical knowledge of auxiliaries
Two separate multifactorial ANOVAs show that – both in NN and in N groups – neither frequency nor BTP scores affected the above-mentioned reading measures (all p’s > .14). To ascertain whether more frequent and more strongly backward collocated constructions predict more target-like judgments, we regressed these factors onto testers’ answers using a linear regression model for NN and N speakers separately. Results show that – for all NN groups and for NSs – when the frequency of construction score increased, so did the probability of target-like judgments (Wald Z = 4.201, p < .001). The same was true for BTP scores (Wald Z = 5.45, p < .001).
6.3 Non-optimal processing of auxiliaries
Table 6 above summarizes reading measures across NN groups and N controls. Overall, NN speakers were two to three times slower than N controls.
Average total reading times (TRT), total reading times on critical zones (TRT-C) and first-pass reading times on critical zones (FP-C) per group (in ms).
| GROUP | TRT | TRT-C | FP-C |
| A | 3789 | 1513 | 574 |
| B | 6035 | 2024 | 690 |
| C | 4410 | 1359 | 603 |
| CNTR | 1873 | 618 | 316 |
According to multifactorial ANOVA, group was a highly significant factor (p***) to explain differences in reading times. Pairwise post hoc comparisons of the means (Tukey’s HSD) show that all NN groups differed from N controls and from each other, except groups A-C in TRT (p = 0.22) and in TRT.C (p = 0.619). First passes of NN groups differed significantly from those of N controls, but not from each other (all p’s > .1).
We have seen that path length (PL) is the sum of the lengths (in pixels) of the segments connecting each fixation with the next one. PL may reflect difficulties in extracting the relevant linguistic information from a written text. PL in NN groups was significantly higher (p***) than in N controls, but Tukey post hoc analysis showed that also the differences among NN groups were significant (p**): groups A and C behaved similarly, while group B was far back (+ 1012 pixel on average).
As far as word-by-word reading is concerned, N speakers’ first passes skipped auxiliaries 67 % of the times, whereas PastPs were seldom skipped (3 %). In contrast, NN subjects skipped auxiliaries and PastPs only 3 % of times: differences among NN groups were not significant. Figure 6 visualizes the n. of fixations on the ROI (auxiliary + past participle) per group.
N. of fixations on the ROI (auxiliary + past participle) across A, B and C groups (CNTRL = N controls).
As to the number of fixations on the whole sentences, group A patterned similarly to the N controls (difference p > .20), while Chinese learners of groups B and C, on average, reported three times more fixations than the N controls (all ps***). Post hoc Tukey showed that also differences among NN groups were significant, with subjects of group B reporting, on average, almost three times more fixations than C and A subjects (all ps**). Finally, to verify whether exposure to the target language, length, and recency of instruction, and (alphabetical vs. non-alphabetical) L1 affected acceptability judgments and reading patterns, we regressed the latter onto the former in a linear regression model and in a multifactorial ANOVA, respectively. The linear regression shows that none of these factors affected the rate of target-like acceptability judgments (Table 7).
Wald Z and ps of independent factors regressed on testers’ responses.
| Factor | Wald Z | p |
| Months in Italy | −0.98 | 0.32 |
| Hours of instruction | −0.08 | 0.93 |
| Recency of instruction | −0.68 | 0.49 |
| Alphabetical L1 | 1.08 | 0.27 |
By contrast, a multifactorial ANOVA shows that an increase in the length of exposure (months spent in Italy), in hours of instructions and recency of instruction inversely correlated with reading times (in total, as well as in critical and spillover zones), the number of fixations, and the number of regressions. While L1 type (i. e., alphabetical vs. non-alphabetical) was never a significant performance predictor, the presence of auxiliation in learners’ L1 positively affected the number of fixations on critical zones (F = 9.97, p = .0001). The first passes in critical zones seemed to be insensitive to all factors. To verify this, five more measures for first passes were added: (i) first passes on auxiliary; (ii) first passes on PastP; (iii) first passes on spillover; (iv) first-pass fixations on auxiliary; (v) first-pass fixations on PastP. None of these first-pass measures was affected by any descriptive independent variables (all p’s > .20).
7 Recap of findings and discussion
7.1 Findings across experimental conditions
There are three main findings concerning the linguistic manipulations of the stimuli:
Our data do not support the hypothesis that beginning L2 learners may be sensitive to a grammatical rule for AuxS concerning the unaccusative/unergative split. Even though the most accurate and fastest readers from group A (alphabetical L1s) behaved like the N controls, in that they processed unergative, peripheral, and mismatching verbs more slowly than unaccusative, core, and type verbs, none of these differences reached statistical significance. Moreover, the target-like vs. non-target-like auxiliary condition was never a significant factor for reading times, number of fixations, and number of regressions. Finally, none of the linguistic manipulations increased the probability of learners successfully spotting the violation in acceptability judgments.
Our data partially support the hypothesis that L2 learners are sensitive to frequency and distribution of auxiliaries in the input: while frequency and BTP score highly correlated with their judgments, they did not affect reading measures.
Our data strongly support the hypothesis that learners’ processing is still non-optimal. All NN subjects read significantly slower than N speakers: they fixated every word, went back, and re-read the sentence many times, as testified by number of regressions and path-length.
7.2 Findings across NN groups
As mentioned, subjects of group A had various alphabetical L1s, while subjects of groups B and C were all Chinese NS. We questioned whether there could have been a natural advantage for left-to-right alphabetical-L1 readers of group A over the Chinese readers of groups B and C. Results showed that the Chinese learners did fixate auxiliaries longer, more frequently, and more times than subjects with alphabetical L1s, but all NN subjects shared the same tendency not to skip auxiliaries. Actually, individual differences among learners entwined along more than one dimension, and learners’ L1s alone could not explain the variance. Chinese subjects belonging to groups B and C patterned inversely with respect to length of exposure and recency of instruction. Instead, group A (alphabetical-L1 subjects) and group C (Chinese subjects still attending Italian courses) were similar throughout all dimensions of variability. In fact, in our data we found that alphabetical L1 was never a factor for either reading measures or acceptability judgments.
As far as the rate of target-like acceptability judgments is concerned, learners of group A were more similar to the N controls than the Chinese learners of groups B and C, even if students of group C were more target-like in their judgments than students of group B (Table 5). With respect to all reading measures (Table 4), the Chinese subjects in group B were significantly behind. These subjects – unlike those in group C – had been attending university courses for two years but they had not taken any language classes since entering university; in addition, they lived in a close community and their interactions with other international students and/or NSs were scarce or null.
In conclusion, all groups were equally sensitive to auxiliary frequency in the input (both raw frequency of construction and BTP) and they did not differ significantly with respect to the rate of target-like acceptability judgments. The difference between alphabetical and non-alphabetical L1 seemed to emerge only in the case of linguistic manipulations, where subjects speaking alphabetical languages behaved like the N controls.
7.3 Conclusions: Non-optimal processing of auxiliaries and precursors of initial L2 learners’ knowledge of auxiliaries
In our study, we advanced the hypothesis that learners could be sensitive to either a grammatical rule or to the frequency of auxiliaries in the input. We also claimed that especially beginning L2 learners could be still unskilled readers of the target language and that this could mask their developing grammatical or statistical knowledge of Italian auxiliaries. We found that our beginner NN subjects were incapable of integrating either the grammatical or statistical information in online sentence processing, which is supposed to be assisted by procedural knowledge of auxiliaries in the NSs’ competence. One can conclude that a frequency-based heuristics is at work, at least when learners are asked to recognize auxiliaries, even though this knowledge is not yet internalized and, therefore, it cannot affect online processing. The most important finding of this study is that all L2 learners – even the more proficient NN subjects – were extremely slow readers: processing a sentence took them twice as long as it took to the N controls. We may hypothesize that our learners were still attuning their processing to the target language and were developing new reading habits, independently of the instantiation of grammatical or statistical representations and of L1 features. Another interesting result is that subjects currently attending language courses (groups A and C) outperformed the non-recent learners of group B, despite the latter group had been living in Italy for at least two years on average and despite B and C shared the same non-alphabetical L1. This could indicate that classroom input and practice too may have had an impact on learners’ developing competence.
In general, our conclusion is that the fatigue of the reading task in the L2 experienced by our beginning NN participants could have at least partially masked both the L2 developing grammar and their sensitivity to statistics. Eye-tracking longitudinal studies where the same subjects serve as their own controls over time could perhaps provide us with proper means to examine procedural knowledge of auxiliaries also through the analysis of reading measures. To our knowledge, such studies have not been carried out yet.
Appendix A List of experimental sentences
| n. | Sentences Stimuli | English Translation | Kind of verb |
| 1 | Luigi ha telefonato a sua mamma con il cellulare | L. called his mother with his cell phone | Type/core |
| 2 | Il bambino è corso perché aveva freddo | The child ran because he was cold | Mismatching |
| 3 | Mohamed è saltato sulla sua motocicletta | M. jumped on his motorbike | Mismatching |
| 4 | Il sub ha emerso in superficie | The diver emerged to the surface | Type/peripheral |
| 5 | Luigi è tremato per il freddo perché era nudo | L. shivered with cold because he was naked | Type/peripheral |
| 6 | Mio fratello è rimasto in città invece di partire | My brother remained in the city instead of leaving | Type/peripheral |
| 7 | Stefano è arrivato alla festa con la sua fidanzata | S. arrived at the party with his girlfriend | Type/core |
| 8 | Sergio ha inciampato mentre camminava | S. stumbled while he was walking | Mismatching |
| 9 | Paolo è giunto alla stazione in orario | P. arrived at the station on time | Type/core |
| 10 | Mario ha continuato a parlare | M. kept on talking | Mismatching |
| 11 | L’uomo è mentito alla sua fidanzata | The man lied to his girlfriend | Type/core |
| 12 | Il tetto è ceduto improvvisamente | The roof yielded suddenly | Type/peripheral |
| 13 | Il motore ha funzionato dopo due tentativi | The engine functioned after two attempts | Mismatching |
| 14 | Mario è vissuto a Roma per lavoro | M. lived in Rome because of his job | Mismatching |
| 15 | Un cellulare è squillato nel silenzio della classe | A cellphone rang in the silence of the classroom | Mismatching |
| 16 | Il petrolio ha mancato a causa della guerra del Golfo | Oil lacked because of the Gulf war | Mismatching |
| 17 | Mario ha vissuto a Roma per lavoro | M. lived in Rome because of his job | Mismatching |
| 18 | L’uomo ha mentito alla sua fidanzata | The man lied to his girlfriend | Type/core |
| 19 | Paolo ha giunto alla stazione in orario | P. arrived at the station on time | Type/core |
| 20 | Mio fratello ha rimasto in città invece di partire | My brother remained in the city instead of leaving | Type/peripheral |
| 21 | Mohamed ha saltato sulla sua motocicletta | M. jumped on his motorbike | Mismatching |
| 22 | Luigi ha tremato per il freddo perché era nudo | L. shivered with cold because he was naked | Type/peripheral |
| 23 | Luigi è telefonato a sua mamma con il cellulare | L. called his mother with his cell phone | Type/core |
| 24 | Il petrolio ha mancato a causa della guerra del Golfo | Oil lacked because of the Gulf war | Mismatching |
| 25 | Il motore è funzionato dopo due tentativi | The engine functioned after two attempts | Mismatching |
| 26 | Il tetto ha ceduto improvvisamente | The roof yielded suddenly | Type/peripheral |
| 27 | Mario è continuato a parlare | M. kept on talking | Mismatching |
| 28 | Stefano ha arrivato alla festa con la sua fidanzata | S. arrived at the party with his girlfriend | Type/core |
| 29 | Il bambino ha corso perché aveva freddo | The child ran because he was cold | Mismatching |
| 30 | Il sub è emerso in superficie | The diver emerged to the surface | Type/peripheral |
| 31 | Un cellulare ha squillato nel silenzio della classe | A cellphone rang in the silence of the classroom | Mismatching |
| 32 | Sergio è inciampato mentre camminava | S. stumbled while he was walking | Mismatching |
Marco Polo and Turandot Program were launched in 2006 by the CRUI, the Conference of Italian University Rectors, in response to an official request by the President of Italy. The concept of the Program was to create a preparatory course of Italian language and culture for Chinese students willing to enroll in any Italian university, to provide them with the knowledge needed to attend classes in L2. See http://marcopolo.unipv.eu/.
References
Alexiadou, Artemis, Elena Agnastopoulou & Martin Everaert. 2004. Introduction. In Artemis Alexiadou, Elena Agnastopoulou & Martin Everaert (eds.), The unaccusativity puzzle. Explorations of the syntax-lexicon interface, 1–21. Oxford: Oxford University Press.10.1093/acprof:oso/9780199257652.001.0001Search in Google Scholar
Avram, Larisa. 1994. Auxiliary configurations in English and Romanian. Revue Roumaine de Linguistique 5(6). 493–510.Search in Google Scholar
Bard, Ellen Gurman, Cheryl Frenck-Mestre & Antonella Sorace. 2010. Processing auxiliary selection with Italian intransitive verbs. Linguistics 48(2). 325–361.10.1515/ling.2010.011Search in Google Scholar
Benincà, Paola & Guglielmo Cinque. 1992. Sur l’ambiguïté structurale des verbes météorologiques en italien. In Liliane Tasmowski & Anne Zribi-Hertz (eds.), Hommages à Nicolas Ruwet: de la musique à la linguistique, 155–162. Ghent: Communication & Cognition.Search in Google Scholar
Bentley, Delia. 2006. Split intransitivity in Italian. Berlin & New York: De Gruyter Mouton.10.1515/9783110896053Search in Google Scholar
Clifton, Charles Jr., Adrian Staub & Keith Rayner. 2007. Eye movements in reading words and sentences. In Roger P. G. van Gompel, Martin H. Fisher, Wayne S. Murray & Robin L. Hill (eds.), Eye movements: A window on mind and brain, 342–371. Amsterdam: Elsevier.10.1016/B978-008044980-7/50017-3Search in Google Scholar
Frenck-Mestre, Cheryl. 2002. An on-line look at sentence processing in the second language. In Roberto R. Heredia & Jeanette Altarriba (eds.), Bilingual sentence processing (Advances in Psychology 134), 217–236. Amsterdam: North-Holland.10.1016/S0166-4115(02)80012-7Search in Google Scholar
Frenck-Mestre, Cheryl. 2005. Ambiguities and anomalies: What can eye-movements and event-related potentials reveal about second language sentence processing? In Judith F. Kroll & Anette M. B. De Groot (eds.), Handbook of bilingualism, 268–281. Oxford: Oxford University Press.Search in Google Scholar
Gries, Stefan. 2015. Statistics for learner corpus research. In Sylviane Granger, Gaëtanelle Gilquin & Fanny Meunier (eds.), The Cambridge handbook of learner corpus research, 159–182. Cambridge: Cambridge University Press.10.1017/CBO9781139649414.008Search in Google Scholar
Inhoff, Albrecht W. & Weimin Liu. 1998. The perceptual span and oculomotor activity during the reading of Chinese sentences. Journal of Experimental Psychology: Human Perception and Performance 24(1). 20–34.10.1037/0096-1523.24.1.20Search in Google Scholar
Irwin, David E. 2004. Fixation location and fixation duration as indicesof cognitive processing. In John M. Henderson & Fernanda Ferreira (eds.), The integration of language, vision, and action: Eye movements and the visual world, 105–134. New York: Psychology Press.Search in Google Scholar
Ježek, Elisabetta & Stefano Rastelli. 2008. Gradiente di inaccusatività e verbi pronominali in italiano L2. In Giuliano Bernini, Lorenzo Spreafico & Ada Valentini (eds.), Competenze lessicali e discorsive nell’acquisizione di lingue seconde, 79–100. Perugia: Guerra.Search in Google Scholar
Keller, Frank & Antonella Sorace. 2003. Gradient auxiliary selection and impersonal passivization in German: An experimental investigation. Journal of Linguistics, 39, 57–108.10.1017/S0022226702001676Search in Google Scholar
Kraš, Tihana. 2010. Unaccusativity in L2 Italian at the lexico-syntax interface. In Michael Iverson, Ivan Ivanov, Tiffany Judy, Jason Rothman, Roumyana Slabakova & Marta Tryzna (eds.), Proceedings of the Mind/Context Divide Workshop on Linguistic Interfaces and Acquisition, 60–71. Somerville, MA: Cascadilla Press.Search in Google Scholar
Levin, Beth & Malka Rappaport Hovav. 1995. Unaccusativity. At the syntax-lexical semantics interface. Cambridge, MA: The MIT Press.Search in Google Scholar
Loporcaro, Michele. 2003. The Unaccusative Hypothesis and participial absolutes in Italian. Perlmutter’s generalization revised. Italian Journal of Linguistics, 15, 199–263.Search in Google Scholar
MacKenzie, Ian. 2006. Unaccusative verbs in Romance languages. Houndmills: Palgrave Macmillan.10.1057/9780230627550Search in Google Scholar
Malec, Wojciech. 2010. On the asymmetry of verb-noun collocations. In Janusz Arabsky & Adam Wojtaszek (eds.), Neurolingusitic and psycholinguistic perspectives on SLA, 126–143. Bristol: Multilingual Matters.10.21832/9781847692429-011Search in Google Scholar
Onnis, Luca, Heidi Waterfall & Shimon Edelman. 2008. Learn locally, act globally: Learning language with variation set cues. Cognition 109(3). 423–430.10.1016/j.cognition.2008.10.004Search in Google Scholar
Oshita, Hiroyuki. 2001. The unaccusative trap in second language acquisition. Studies in Second Language Acquisition 23(2). 279–304.10.1017/S0272263101002078Search in Google Scholar
Pickering, Martin, Steven Frisson, Brian McErlee & Matthew J. Traxler. 2004. Eye movement and semantic composition. In Manuel Carreiras & Charles Clifton, Jr. (eds.) The on-line study of sentence comprehension: ERP, eye-tracking and beyond, 134–156. New York: Psychology Press.Search in Google Scholar
Randall, Janet. 2007. Parameterized auxiliary selection: A fine-grained interaction of features and linking rules. In Raúl Aranovich (ed.), Split auxiliary systems, 207–235. Amsterdam: Benjamins.10.1075/tsl.69.10ranSearch in Google Scholar
Rayner, Keith & Sara C. Sereno. 1994. Eye movements in reading. In Morton Anne Gernsbacher (ed.), Handbook of psycholinguistics, 57–82. San Diego, CA: Academic Press.Search in Google Scholar
Rosemeyer, Malte. 2014. Auxiliary selection in Spanish. Gradience, gradualness, and conservation. Amsterdam: Benjamins.10.1075/slcs.155Search in Google Scholar
Shillcock, R. 2007. Eye movements and visual world recognition. In M. Gareth Gaskell (ed.) The Oxford handbook of psycholinguistics, 89–105. Oxford: Oxford University Press.10.1093/oxfordhb/9780198568971.013.0006Search in Google Scholar
Sorace, Antonella. 1993. Incomplete vs. divergent representations of unaccusativity in native and non-native Italian grammars of Italian. Second Language Research 9. 22–47.10.1177/026765839300900102Search in Google Scholar
Sorace, Antonella. 2000. Gradients in auxiliary selection with intransitive verbs. Language 76(4). 859–890.10.2307/417202Search in Google Scholar
Sorace, Antonella. 2004. Gradience at the lexico-syntax interface: Evidence from auxiliary selection and implications for unaccusativity. In Artemis Alexiadou, Elena Agnastopoulou & Martin Everaert (eds.), The unaccusativity puzzle. Explorations of the syntax-lexicon interface, 243–268. Oxford: Oxford University Press.10.1093/acprof:oso/9780199257652.003.0010Search in Google Scholar
Sorace, Antonella. 2015. The cognitive complexity of auxiliary selection: From processing to grammaticality judgements. In Rolf Kailuweit & Malte Rosemeyer (eds.), Auxiliary selection revisited: Gradience and gradualness, 23–41. Berlin & New York: De Gruyter.10.1515/9783110348866-002Search in Google Scholar
Staub, Adrian & Keith Rayner. 2007. Eye movements and on-line comprehension processes. In M. Gareth Gaskell (ed.), The Oxford handbook of psycholinguistics, 327–342. Oxford: Oxford University Press.10.1093/oxfordhb/9780198568971.013.0019Search in Google Scholar
Van Hout, Angelik. 2004. Unaccusativity as telicity checking. In Artemis Alexiadou, Elena Agnastopoulou & Martin Everaert (eds.), The unaccusativity puzzle. Explorations of the syntax-lexicon interface, 60–83. Oxford: Oxford University Press.10.1093/acprof:oso/9780199257652.003.0003Search in Google Scholar
VanPatten, Bill. 2007. Input processing in adult SLA. In Bill VanPatten & Jessica Williams (eds.), Theories in second language acquisition, 115–135. Mahwah, NJ: Erlbaum.10.4324/9781410615299Search in Google Scholar
© 2022 the author(s), published by De Gruyter
This work is licensed under the Creative Commons Attribution 4.0 International License.
