«Learning a Gendered Language: L2 Acquisition and Relative Processing Costs of Spanish Grammatical Gender and Number Alexander M. Hirsch Yale ...»
linguistic features such as Spanish gender, perhaps just as capably as heritage speakers. Sagarra & Herschensohn (2010) continue this line of investigation by examining L2 learners’ processing of gender and number agreement via various tasks involving Spanish nouns and their adjectival complements, which must agree in gender and number in Spanish. After testing native speakers and English-L1 Spanish-L2 beginning and intermediate learners, the investigators found that, as predicted, beginning L2 learners performed poorly on on-line and off-line discrimination tasks of gender and number agreement, native speakers performed at ceiling levels, and intermediate L2 learners performed somewhere in between these two groups. Furthermore, all three groups’ off-line performances indicated equal performance on gender mismatches and number mismatches. While the lack of a significant performance difference between gender and number concord/discord in the beginning learners and the native speakers can be explained by floor and ceiling effects, respectively, the performance of the intermediate learners cannot be explained as such. Indeed, the intermediate L2 learners identified number mismatches more accurately than gender mismatches in the off-line grammatical judgment task (natives again performed at ceiling levels). Furthermore, working memory capacity correlated to reaction times to gender disagreement but not to number
The investigators reconcile the differing results of the on-line and off-line tasks by stating that their (on-line) self-paced reading task is not sensitive enough to discriminate between processing performance based on such fine distinctions. This explanation is not altogether satisfying. A better interpretation might invoke the relatively low proficiency of these intermediate L2 learners, a group which has been shown, as mentioned above, to perform significantly differently from more advanced learners and from near-natives, possibly to an exponential extent.
Regardless of the explanation of the lack of demarcation between gender and number in the on-line task, the fact that Sagarra & Herschensohn do find any distinction among beginning and intermediate learners between gender and number concordance suggests that such a processing distinction may in fact exist. A subsequent study by the same authors (Sagarra & Herschensohn 2011) on gender agreement processing of inanimate versus animate nouns showed that intermediate learners tend to behave more like native speakers than like beginners. That is, they exhibit longer reaction times (and therefore more processing difficulty) for animate than inanimate nouns and for gender discord than concord. If the results of this second study can be extended to those of the first, they suggest that the test itself may not have been the source of the lack of differentiation between gender and number, especially considering that the same methodology was used in both studies. Rather, it seems that some other issue was at play.
Perhaps the low proficiency of the intermediate learners allows for a distinction to be made between animacy and inanimacy but not between gender and number and such a distinction will emerge in higher-proficiency speakers. Regardless of the specific
carry different processing burdens is one of the major inquiries that the study at hand seeks to address and one that this investigator feels has not been adequately engaged in the past.
4B. Neurological data. Electrophysiological data supports the behavioral findings that highly advanced or near-native L2 speakers can achieve similar proficiency to that of native speakers. Gillon Dowens, et al. (2010) conducted an ERP study of L1-English, nearnative L2-Spanish speakers in which subjects (and native Spanish controls) were exposed to sentences containing determiner-noun and noun-adjective number and gender mismatches either within a phrase or across a phrase boundary while their brain activity was monitored via EEG. Compared to the native controls, the L2 speakers exhibited similar patterns of brain activity. Intra-phrase violations elicited the same pattern of an early left anterior negativity (LAN) followed by a P600 response from the L2 speakers as from the natives. When a violation occurred that spanned across two phrases, however, no LAN effect was elicited. The investigators ascribe this difference to the increased processing demands that come with retaining information of long-distance dependencies, which would presumably reduce a speaker’s ability to react to these kinds of violations. Returning to the intra-phrase responses, while the native speakers showed identical responses to gender as to number violations, the L2 speakers showed qualitative differences between the response curves for intra-phrase gender versus number violations, suggesting that these constructions are not being processed in a completely nativelike manner.
Although the observations that this study makes on the nature of second language processing are clear, the question remains as to whether L2 transfer substantially
study by Gillon Dowens, et al. (2011) seeks to answer this question. Using the same methodology as Gillon Dowens, et al. (2010)’s study, the experimenters tested L2 Spanish learners whose native language was Chinese, rather than English, against native Spanish speakers, again using ERP techniques. Unlike English and Spanish, Chinese contains neither gender nor number agreement, 4 so there is no chance that a difference in processing between gender and number agreement could be conditioned by L1 transfer. ERP recordings showed a P600 effect in response to both gender and number violations, with no reliable difference between the two conditions. Unlike the English natives described above, the Chinese natives showed no LAN effect even when a violation occurred within a phrase. The prevailing explanation of the LAN is that it corresponds to more automatic stages of morphosyntactic processing, while the P600 is thought to be related to more controlled processing of higher syntactic features (Friederici, Steinhauer, & Pfeifer 2002;
Friederici 2001). Therefore, Gillon Dowens, et al. (2011), suggest that the lack of any LAN response from the Chinese natives could be due to their somewhat lower proficiency in Spanish—they were living in a Chinese-speaking environment at the time of the study and were not immersed in the L2—compared to the English natives, who were living in a Spanish-speaking country and were speaking in Spanish on a daily basis. Unfortunately, such a complication could be a major confounding factor for this type of research and it is therefore difficult to take any solid conclusions from the lack of a LAN. Even so, the fact that the Chinese natives did not show a difference in their P600 responses to gender versus number violations suggests that L1 transfer may indeed have played a role in the English natives’ differential responses observed in the earlier study.
In fact, Chinese completely lacks any inflection for gender, number, person, or case. (Gillon Dowens, et al.
Although the abovementioned work by Sagarra & Herschensohn and Gillon Dowens, et al., is a good first step to describing the behavior of advanced and near-native L2 speakers in response to gender and number violations, these authors do not attempt to provide explanations for this behavior. The following section continues in this direction, exploring the underlying cognitive architecture supporting nativelike parsing of L2 gender and number processing.
5. Foundations of Gender and Number Processing The studies above begin to provide a clearer picture of the possible differences or similarities in processing gender versus number inflection in Spanish. However, these studies tend towards description of behavior and cerebral activity, while leaving out any deeper syntactic and cognitive basis for these observations. The studies discussed below, on the other hand, delve deeper into the cognitive underpinnings of observed distinctions between gender and number, as well as between native and nonnative processing of these features. Although the data from the two studies headed by Gillon Dowens, discussed above, seems fairly clear-cut and cogent at first blush, other studies have come to different conclusions about processing of gender and number marking. For syntactic reasons that are largely beyond the scope of this paper, it has been argued at least as early as 1993 (by Ritter) that gender and number may be fundamentally different morphosyntactic constructs. This claim can be compared to the hypothesis described in section 1D, above.
Perhaps gender and number fall on different sides of the division between “words” and “rules,” to use Ullman’s terminology. Gender, an intrinsic lexical feature, would belong to the “words” subset, while number, a feature layered onto the existing lexical item, could be
is easy to extend this difference to processing as well and to hypothesize that these two features (for lack of a better word) may be processed differently from one another, even in native speakers. Preliminary behavioral studies seem to bear this out: in one study of gender and number agreement errors by native speakers (Vigliocco, Butterworth & Garrett 1996), number errors far outnumbered gender errors. If the lower instance of gender agreement errors can be taken to be suggestive of a deeper integration of gender than number, these results imply that gender may be an intrinsic part of a lexically-stored word or lemma, while number is not (and perhaps is instead derived by the application of a rule or some other mechanism).
The claim that grammatical gender is inherent to the noun while number is rulederived has interesting implications. In a study of reaction times to gender versus number incongruencies, Faussart, Jakubowitz, & Costes (1999) found that gender agreement violations elicited a longer reaction time than number violations. The authors explain this finding by positing that the reanalysis of a morphosyntactic feature that is processed earlier in the lexical retrieval routine requires that more of the routine be re-run, while a later feature like number only requires that the final step of the routine be repeated.
Therefore, the reanalysis process, measured in this study via reaction time, is shorter for number agreement violations than those for gender, even though gender is processed at an earlier and possibly more innate or instinctive stage in the parsing process than number.
Without physiological evidence like that obtained through the use of ERP techniques, however, such hypotheses are mostly limited to a broad psycholinguistic conception of the phenomenon at hand. Luckily, many ERP studies on the subject have been conducted in
Certain ERP studies in the past have sought to answer exactly the question of whether gender or number is comparatively easier to process than the other. One such study, conducted by Barber & Carreiras (2005), examined gender and number agreement in Spanish both between determiners and nouns and between nouns and adjectives.
Although many studies of ERP responses to either gender or number agreement individually have been conducted in the past, no study had previously looked at gender and number agreement within the same study while also using a language like Spanish that contains both markers in a grammatical framework (compared to a language like English in which gender is determined only by real-world semantic value and there is no robust gender agreement). Therefore, the Barber & Carreiras study was effectively the first to directly compare processing of gender versus number mismatches via ERP.
The study consisted of two separate but related experiments. In the first, native Spanish-speaking participants were given two-word pairs, either an article and a noun or a noun and an adjective. In the second, a separate group of native Spanish-speakers were shown a sentence with an agreement violation either at the beginning of the sentence— between the article and the subject noun—or in the middle of the sentence—between the subject noun and the adjective. Filler sentences without agreement errors were also included. In both experiments, participants were then asked specifically to judge the grammaticality of the word pair or sentence, though the subjects were not instructed to complete the task within any kind of time limit. The subjects’ cognitive activity was monitored throughout via ERP recording.
In accordance with the research described above, Barber & Carreiras’s results
number violations. Specifically, ERP results showed that both types of agreement violation elicited a P3 component (a positivity between about 200 and 500 ms) in response to the word pairs, and a P600 component in response to the sentences. The two types of violations differed qualitatively, however: gender violations elicited a later P3 and P600 than the equivalent number violations. These differences in number versus gender processing occurred in late time windows, during which parsing reanalysis is generally thought to occur. This finding can be explained if it is assumed that gender is a lexical feature as well as a syntactic feature (because syntactically connected words must agree in gender), while number is solely syntactic, as it is layered on top of preexisting lexical items after retrieval. Therefore, reanalysis of a gender agreement violation will have to check not only syntactic structure-building but also lexical access, while detection of a number violation would only require syntactic structure reanalysis.
The above results are unlike those of previous ERP-based studies which found no dissociation between processing of gender and which seemed to contradict behavioral studies showing robust reaction-time differences. Instead, Barber & Carreiras’s findings are consistent with previous behavioral results, showing delayed late-stage processing of gender compared to number violations.