«Learning a Gendered Language: L2 Acquisition and Relative Processing Costs of Spanish Grammatical Gender and Number Alexander M. Hirsch Yale ...»
1. Experimental Data and Results 1A. Reactions to the Word Preceding the Noun First, the results for the word immediately preceding the noun did not show any meaningful differences between conditions or groups, nor did they show any interactions between the two. These results largely follow the predicted outcomes: there should be no significant effect produced by the word preceding the noun since this word was not selected specifically for any reason.
Indeed, this word was often a determiner or a number word, which are very common and should not elicit any specific response. It was predicted, however, that L1 speakers would have lower reaction times overall than L2 speakers given that L1 speakers are known to be able to read faster and more automatically than L2 speakers (Sagarra & Herschensohn 2010). This type of effect was not observed in these results. It is unclear whether this was due to the actual nonexistence of this effect in the studied populations or to the small
is possible that with more subjects, there would be a clear difference overall in mean RTs between the L1 and L2 groups to the word preceding the noun (as well as the noun and the following word).
1B. Reactions to the Experimental Noun As predicted, the data collected on RT to the experimental noun differed from those on RT to the preceding word. Namely, significant results were observed for the L2 population and results approaching significance were observed for the L1 population. Main effects for both group and condition suggest that both variables play a role in subjects’ comprehension of these types of sentences. For Spanish natives (the L1 group), although individual T-tests did not show significant differences between gender violations and the agreement control sentences or number violations and agreement controls, the fact that the results approached significance suggest that a larger sample size might have yielded reaction times to gender and number violations that were statistically significant compared to those to agreement. The L2 group did show this kind of differentiation with respect to number violations compared to agreement, though not to gender violations. This result can be explained by the fact that, because English has number but not gender agreement between determiners and nouns, the L2 speakers were sensitive to number violations but not to gender violations in the narrow window immediately after the experimental noun. This possibility will be explored further in the General Discussion below. Native speakers, on the other hand, only reacted at time scales approaching significance. Although it is possible that a larger sample population would have produced more robust RT data, it can only be conjectured that natives would react significantly slower to gender as well as number mismatches. That native speakers
p value was calculated for the reaction times to gender compared to number mismatches— suggests that, in this early time window, native speakers react quite the same to gender and number errors. If this is the case, native speakers would seem to lack the separation in their responses to gender and number mismatches that has been proposed by Faussart, Jakubowitz, & Costes (1999); again, this possibility will be examined further below.
However, because significance was not achieved, further experimentation is necessary to confirm or deny this possibility.
1C. Reactions to the Word Following the Noun Looking at the results for RT to the word following the noun, a main effect for condition suggests that speakers remained sensitive to agreement beyond the window immediately after the noun, while a main effect for group suggests that the sustained responses of the two groups were still divergent from one another. Post hoc, the results of the various T-tests give a somewhat different picture from the noun-RT data discussed above. In response to the word following the noun, L1 speakers responded significantly slower only to number violations but not gender violations (although gender RTs approached significance). This result suggests that number reanalysis by the native speakers occurs more quickly that gender reanalysis.
Furthermore, it accords with work by Barber & Carreiras (2005) and Faussart, Jakubowitz, & Costes (1999) (described in the literature review, above): if gender is both a lexical and syntactic feature, it would be built into the syntax at the lexical retrieval stage, while number, a syntactic but not lexical feature, is attached to words already built into the syntactic structure. Therefore, number reanalysis would be a shorter process than gender reanalysis, leading to a less-delayed reaction time for number than gender. This hypothesis
obtained when comparing gender and number directly, the available results suggest that such a possibility should be explored in greater detail, as a statistically significant difference between the two would further highlight the delayed processing of gender errors compared to number.
L2 speakers, on the other hand, showed significant differences in reaction time for both gender and number as compared to agreement. Compared to their reactions to the noun itself, this result suggests that L2 speakers indeed exhibit delayed processing for gender alone, even when processing of number violations is not delayed (significant RT differences due to mismatch reanalysis were observed between number violations and agreement at the noun). If it were true that L2 speakers could not acquire a nativelike gender processing ability, these speakers would not show slowed reaction times in response to gender errors but rather would only exhibit such slowing in response to number errors. Instead, the L2 speakers reacted more slowly to both mismatch conditions than to agreement, suggesting that both gender and number errors provoked a nativelike response from the L2 speakers at the word following the noun. The fact that these speakers were observed to react to gender mismatches in such a nativelike way suggests that gender may be acquirable by L2 speakers past the superficial level that the various Critical Period hypotheses suggest.
2. General Discussion As discussed in the introduction to this paper, although the behavioral data on native speaker processing of gender and number in Spanish does not always show a clear distinction between the two features, neurophysiological data from Barber & Carreiras (2005) and others show a concrete qualitative difference in processing of number versus
gender agreement production has been shown to be more automatic than number, while gender mismatch comprehension and reanalysis has been shown to be delayed compared to number. As mentioned above, data collected on the RTs of L1 speakers to gender and number mismatches was aligned with the conclusions from the literature, though the results obtained were not adequately significant to confirm or deny any difference between responses to gender versus number mismatches. Luckily, the results for the experimental group – the L2 learners – were more conclusive.
Because the L2 Spanish speakers were all native speakers of English, it was predicted that number processing would be more automatic (faster) than gender, which would be delayed. As mentioned in the introduction, English has gender marking on neither common nouns nor determiners, and there is no syntactic gender agreement whatsoever. Members of these word classes often do carry number marking, however, and number must agree between nouns and certain types of determiners (e.g. this/these, that/those, a/some, etc.). Therefore, English natives should be familiar with number marking and agreement as a functional concept to a greater extent than gender. The expected consequences of this observation vary depending on adherence to Clahsen & Felser’s Shallow Structure Hypothesis. According to the SSH, transfer does not play a role in L2 acquisition, although as described in the introduction, subsequent work by Sorace (2006) and others suggests that this position is too strong and that lower L2 proficiency may actually encourage parsing via transfer up to a certain point. The results of the present experiment suggest that L2 learners employ at least some degree of transfer in their parsing of number mismatches. L2 learners exhibited a significantly slower reaction time
noun, while native speakers showed approximately similar reactions to gender and number mismatches following the noun. As a result, it is likely that L2 learners did utilize transfer to a certain extent. If L2 learners were not using their English-derived facility with number agreement, they would be expected to react equivalently to gender and number mismatches. Instead, a statistical comparison between these speakers’ responses to gender and number mismatches was nearly strong enough to justify the statement that L2 learners process number errors more natively than gender, suggesting that a larger sample size may have elicited significant separation between their reactions to the two features. The fact that L2 learners reacted to both gender and number mismatches with significantly different RTs from agreement at the position of the word after the noun suggests that at this later time window, gender mismatch processing has caught up with number. The insignificant difference between the two reinforces the idea that gender mismatch processing is present but delayed in L2 learners.
Although the SSH explains many situations involving L2 processing, it is possible that performance differences between the two features could be attributed to the features’ differing concreteness and relation to reality rather than pure syntax. In both English and Spanish, number marking is directly related to the reality that the sentence describes: a plural noun describes a group of objects or people, while a singular noun describes only one object (or collection of objects, in the case of words like “set,” “family,” etc.). Gender, on the other hand, is completely abstract except when describing animate objects (people or animals). Furthermore, nouns describing animals often do not correspond in any real way to an animal’s physical sex (e.g. el perro, the-M dog-M vs. la rana, the-F frog-F), and even
witness-M, which can refer to both male and female witnesses, or la victima, the-F victim-F, which can refer to both male and female victims). 5 Therefore, conceptually speaking, a concrete feature like number is easier to reconcile with the subject matter at hand than gender, a feature that is largely abstract. Furthermore, Corbett (1991) notes that, while gender is abstract, it is also invariable for a given noun, except in cases where the noun’s gender reflects the gender of the person described by the noun. Number, on the other hand, is variable and may take one of two forms in both English and Spanish. Therefore, L1English Spanish L2 speakers might have greater ease with the processing of number than gender because of the processing cost demanded by featural (gender) retrieval. Native speakers’ lexicons, on the other hand, would not treat gender as a feature on top of the noun but instead would contain lexical items in which the noun and its gender are tightly integrated and largely inseparable in terms of recall. Therefore, native speakers’ lexical retrieval processes would recall both gender and noun together and automatically, while the decision of whether to mark the noun as singular or plural (or retrieve the singular or plural form) would require an extra processing step not required by gender recall, allowing faster production of gender than number. Error reanalysis would then proceed by checking agreement of the disagreeing feature. Number mismatches would provoke a check of number, a feature applied to the noun with the same feature on the determiner, which would not require a full lexical retrieval process. Gender mismatches, on the other hand, would require the noun itself to be checked with the determiner’s gender, instantiating lexical retrieval and ultimately taking longer than number mismatch reanalysis. Therefore,
though gender would be produced more automatically than number, gender mismatch reanalysis would take longer than the equivalent process for number.
Although the L2 learner data does not match up exactly with the SSH, it supports the above hypothesis; namely, gender is retrieved and built onto the syntax before number and therefore gender error reanalysis educes a longer reaction time than the equivalent process for number. However, unlike native speakers, L2 learners exhibit signs of transfer, as results of studies by Gillon Dowens, et al. (2010 and 2011) suggest. Even when processing of mismatches in both gender and number is quite nativelike, L2 learners still exhibit a delay in the onset of slowed reaction time to gender but not number errors, possibly due to the influence of their previous experience with number agreement in English. One major limitation of this study was the failure to distinguish between L2 speakers based on proficiency. Though all participants were of relatively high proficiency, likely equivalent to the “advanced” level described by many other investigators of second language acquisition, this study was not able to separate L2 learners into discrete, statistically robust groups based on proficiency because proficiency level was based on the results of a questionnaire rather than a more standardized method. Therefore, as noted by Jegerski, Van Patten, & Keating (2011), it is quite possible that L2 learners at this level could attain a more nativelike level of gender mismatch processing that would bring their performance more into line with the native speakers. In addition, although the task at hand relied on online performance and did not directly test the L2 learners’ grammars as others have done, such a consideration should not affect the reliability of the results, as Jegerski, et al., found that shortcomings in subjects’ performance on online tasks does not entail
adequately advanced that deficiencies in their grammars would not have an effect on their performance, the responses to the background questionnaire showed that participants were sufficiently proficient for the requirements of the supplied task.