«Learning a Gendered Language: L2 Acquisition and Relative Processing Costs of Spanish Grammatical Gender and Number Alexander M. Hirsch Yale ...»
Interpretation of these findings with respect to L2 processing of gender versus number could go one of two ways. If it is assumed that L2 learners assign gender to the lexical entries for each word in their lexicon, then L2 processing should follow the same pattern as native speakers, with gender violations eliciting a longer response than number violations. However, if it is instead assumed that L2 learners without native syntactic
could explain Gillon Dowens, et al. (2011)’s L1-Chinese results, discussed above, it is feasible that L2 learners treat gender like number in that it is applied after lexical retrieval, perhaps through some other retrieval method. Even so, the argument could be made that whatever lexically-external means is used to retrieve gender information would, upon reanalysis, exact a similar processing cost to lexical re-retrieval in native speakers.
In the final introductory section below, the current study is discussed both with reference to the research examined above and with a mind to the detailed methods, results, and conclusions that follow.
6. The Current Study The current study approaches the question of gender vs. number processing differentiation from a behavioral perspective. Using the self-paced reading methodology, participants’ reaction times to gender and number violations between nouns and adjectives within a sentential framework were used to evaluate Spanish-L2 learners’ facility for processing the two markers compared to native controls. Based on the research presented above, we predict that native speakers will exhibit longer reaction times to gender errors than to those for number, both of which will be longer than the reaction times to the errorfree control sentences. L2 learners with English as their native language, on the other hand, are predicted to react more slowly to number than gender errors. Because the L2 learners are assumed to already be familiar with number agreement through transfer from their L1, they will be sensitive to number agreement in their L2 as well (according to any of the various theories of L1 transfer). Being sensitive to number agreement, they will exhibit longer reaction times when number agreement is violated, and delayed reactions in
to gender errors, on the other hand, L2 learners will not be sensitive (or as sensitive) to gender agreement compared to number. This attenuated sensitivity could surface as either limited differentiation between reaction times to gender disagreement compared to control, or it could surface as a delayed effect, in which case the slower reaction time would still be elicited by gender errors, but instead of emerging immediately following the disagreeing word, it would consistently emerge later in the sentence.
Evidence from the study by Gillon Dowens, et al. (2010), among others, suggests that L2 learners process gender qualitatively differently from number. Furthermore, evidence from Keating (2009) and others suggests that while advanced and near-native L2 speakers are often able to process syntactic constructions without an L1 analogue with nearly the same facility as natives when no additional processing cost exists (e.g. long distance dependencies for gender, number, case, etc.), their performance falls off with the addition of these types of external costs. This result suggests that near-natives may indeed possess similar processing capabilities to natives but that their processing gets delayed and bogged down when extra requirements are introduced. In a situation such as the one under study here, where L2 learners’ reactions to errors are judged in an on-line scenario, while processing of a feature present in the L1, such as number agreement, may be processed in a nativelike manner, processing of a feature such as gender agreement that does not benefit from transfer effects may be delayed compared to number. Therefore, it is predicted that, like native speakers, L2 learners will still exhibit slower reaction times to gender errors than to control sentences. However, these reaction times will surface later on, perhaps after the word following the error rather than after the error itself. On the other hand, L2
they will process gender errors, resulting in reaction time delays immediately after the error. Finally, as general trend, L2 learners may exhibit slower reading times overall than native speakers, as observed by Sagarra & Herschensohn (2010). The precise methodology utilized here will be described in detail in the next section.
Participants Participants were drawn from the student population of two Spanish summer school classes at a northeastern United States university, as well as from students in other departments at the university. All participants were native speakers of either American English or Spanish. Participants were divided into two groups based on length of Spanish exposure, either in a formal or naturalistic setting: advanced (greater than three semesters of Spanish, some Spanish immersion), and native (Spanish immersion from before age 10, continued Spanish use). Division of participants was similar to that of Sagarra & Herschensohn (2010). Data on which the division of participants was based was obtained through a short language background questionnaire administered prior to the experiment.
Participants were compensated $5 in the form of an Amazon gift card.
Stimuli Language Background Questionnaire Prior to the experimental task, subjects were asked to complete a questionnaire detailing their linguistic background in Spanish. Subjects were asked about language nativity, age of acquisition, speaking frequency (i.e. hours per week, approximately), living experience in a
were anonymous and were tied to experimental responses using subject codes that were not connected to any personally identifiable data.
Self-Paced Reading Task The present study utilized a self-paced reading task in which participants were asked to read sentences and respond to the prompt, “Does this sentence accurately describe the picture?” given a sample picture. The experimental sentences were either grammatical or contained errors in determiner-noun agreement in either gender (masculine/feminine) or number (singular/plural). Sentences that were semantically anomalous given the picture
were also included as filler sentences. For example:
1. *Hay un hombre a la izquierda que lleva un camisa con estrellas blancas.
“There is a man on the left wearing aMASC.SG shirtFEM.SG with white stars.” 2. *Dos soldado vestidos de verde detienen un hombre delante del coche.
“Two soldierSG dressed in green are arresting a man in front of the car.” 3. #El vendedor de cuadras a la izquierda lleva una camisa azul.
“The picture vendor on the left is wearing a blue shirt.” – anomalous given that the vendor pictured is wearing a pink shirt
4. Hay cinco hombres sentados bajo del toldo de una tienda.
“There are five men seated below the awning of a store.” Of the 80 sentences presented, 40 were semantically relevant to the picture and 40 were semantically anomalous. In addition, 40 of the 80 were grammatical and 40 were ungrammatical. Of the ungrammatical sentences, 20 contained gender mismatches and 20 contained number mismatches. Ungrammatical sentences were distributed throughout the semantically-relevant and semantically-anomalous groups. Participants were instructed to
were also informed that the sentences might contain errors but that they should only answer the question of whether the sentence corresponded to the picture based on the information contained in the sentences.
After each sentence, a picture was displayed on the screen and, as mentioned above, participants were asked to identify whether the information contained in the preceding sentence matched the picture. Question response accuracy rate was used solely to determine whether the participants were in fact attentive to the stimulus at hand.
The experimental paradigm for self-paced reading was a non-cumulative Moving Windows task, as in Just, et al. (1982), and was run using the Paradigm software (Perception Research, 2012). Sentences were displayed onscreen as a series of dotted lines, each representing a single word; the end of the sentence was signaled by a period.
Participants were asked to press a button on the computer keyboard, which would show the first word of the sentence and then each subsequent word, hiding any previous words.
After the last word of the sentence was displayed, a picture and then the question “Does the sentence accurately describe the picture?” were displayed, followed by a prompt for yes or no. Data was collected for time between button presses as well as for question response accuracy.
Self-Paced Reading Task The study followed a mixed effect model, with one between-subjects factor, group (L1 versus L2 speakers), and one within-subjects factor, condition (gender violation,
Results were analyzed via three two-factor ANOVAs of reaction time (RT) with a 3 × 2 factorial design (determiner agreement × group): one for the word immediately preceding the noun, one for the noun, and one for the word following the noun. Descriptive statistics for the self-paced reading trials are presented in Table 1, above. RT preceding the noun was analyzed to ensure baseline stimulation preceding the experimental conditions. RT to the word following the noun was analyzed to determine whether subjects experienced delayed or prolonged processing of the experimental conditions. Late processing effects are important because differences between gender and number processing often have a delayed onset (Barber & Carreiras 2005), and because L2 processing is generally delayed compared to L1 (Weber-Fox & Neville 1996). In addition, performance on comprehension questions was analyzed to ensure higher-than-chance accuracy, which would indicate that subjects were paying adequate attention to the sentences at hand.
Reactions to the Word Preceding the Noun The ANOVA performed on the word
0.05) or condition (F(1,554) = 0.652, p 0.05), and no interaction was found between group and condition (F(2,554) = 0.575, p 0.05). A post-hoc T-test between L1 and L2 speakers did not find a significant difference in reading time of the kind found by Sagarra & Herschensohn (2010) and others.
Reactions to the Experimental Noun The ANOVA performed on the noun itself showed main effects for both group (F(1,554) = 13.82, p 0.05) and condition (F(1,554) = 3.75, p 0.05) but no interaction between group and condition (F(2,554) = 0.80, p 0.05).
Post-hoc one-way ANOVAs performed within each group but across conditions showed no main effect for the L1 group (F(2,237) = 1.86, p = 0.158), while the L2 group’s results approached a significant main effect (F(2,317) = 2.56, p = 0.079). In addition, post-hoc Ttests found a significant difference for number violation RT compared to agreement for L2 speakers (p = 0.044), but not for gender (p = 0.636). Results approached significance for number (p = 0.118) and gender (p = 0.123) violation RTs compared to agreement for L1 speakers, suggesting that a larger sample size might have revealed a significant result in this domain. Furthermore, a T-test comparing gender violation RTs to number violation RTs showed no significant result with an extremely high p value for L1 speakers (p = 0.973) and results approaching significance for L2 speakers (p = 0.106).
Reactions to the Word Following the Noun The ANOVA performed on the word immediately following the noun showed a main effect for group (F(1,540) = 10.72, p 0.05) and condition (F(1,540) = 8.51, p 0.05) but no interaction between group and condition (F(540) = 0.27, p 0.05). Post-hoc T-tests showed significant differences between number violation and agreement RTs for L1 speakers (p = 0.028) and results
(p = 0.118). Significant differences were observed in the results of T-tests between both gender (p = 0.037) and number (p = 0.005) violations and agreement for L2 speakers.
Gender vs. number comparisons of RT to the word following the noun did not yield significant results for either L1 or L2 speakers (L1: p = 0.342; L2: p = 0.316).
Note: n = 3 for L1 speakers, n = 4 for L2 speakers; K = 20 for gender and number violation and K = 40 for agreement.
As for the comprehension questions, all participants performed significantly above chance, as shown by individual binomial tests of each subject’s data (p 0.05 for all subjects). Data broken down by sentence type and speaker group is presented in Table 2, above. The average score was 73.75% correct, or 59 out of 80 questions correct. However, the fact that six of the 80 questions were answered incorrectly by every subject and one more was answered incorrectly by all but one subject likely points to bad sentence construction for the given picture associated with those questions than to overall poor subject performance. A T-test comparing the results of the comprehension questions by the L1 speakers versus the L2 speakers did not show a significant difference between the two
Background Questionnaire Questionnaire results indicated that three of the seven participants were native speakers, while four were advanced L2 learners. L2 speakers had studied Spanish for between two and 20 years, with a mean length of 10 years.
The results of the self-paced reading task have various implications. I will begin by discussing the data from each sentence position independently of the others. Then, I will discuss the results generally with regard to their implications for gender acquisition by L2 speakers specifically and second language acquisition as a whole.