The purpose of the present study is to investigate whether or not the notional number attraction phenomenon is evident during L2 learners’ online processing of subject–verb agreement. L2 sentence processing studies have frequently reported the L2 learners’ online insensitivity to number mismatches, and it has been claimed that L2 learners have difficulty in representing grammatical number features, or they have a tendency to fail to access the number features in online tasks. L2 learners’ sensitivity to
“notional number”, however, has not been considered by researchers. Hence, this research conducted a self-paced reading study targeting highly proficient Japanese learners of English (N = 28). The participants read the three types of sentences; (a) control (e.g., everyone in the room was/were…), (b) notional attraction (e.g., everyone in the team was/were…), and (c) grammatical attraction (e.g., everyone in the halls was/ were…). In order to examine the participants’ sensitivity to number attractions, the differences in the reading times between the number matched and the mismatched verbs (was/were) were compared among the attraction types. The observed reading time differences in the grammatical and notional attraction condition were different than in the control condition, indicating that both attraction phenomena were evident. L2
Graduate School, Nagoya University Japan Society for the Promotion of Science
Graduate School, Nagoya University Japan Society for the Promotion of Science
The Notional Number Attraction in English as a
Foreign Language: A Self-Paced Reading Study
Graduate School, Nagoya University
learners’ imperfect representation of number features, which we call representation vulnerability, was discussed.
Keywords: sentence processing, number agreement, notional number
1.1. Background of the Present Study
Unlike sentence processing in the first language, that of second/foreign language (L2) learners exhibits great variety among individuals (e.g., Clahsen & Felser, 2006). The difference might be explained by the L2 learners’ first language, which is one of the most important factors affecting L2 acquisition and sentence processing. In addition to the L1 influence, developmental factors are also worth considering because they may make the system underlying L2 learners’ sentence processing unsteady, which is not the case for adult native speakers. Individual traits such as working memory capacity, age, learning environment, and numerous other psychological traits can also be important factors to explain the variance (e.g., Robert & Meyer, 2012). More importantly, the strongest factors that affect the way L2 learners process the input may be the grammatical structure and linguistically related issues. In other words, it is quite likely that L2 learners, particularly Japanese learners of English, struggle with certain grammatical structures such as number agreement (detailed below) and English articles. What is more, such linguistic factors interact with task-related factors in a highly complex manner (Task–Structure Interaction; e.g., Kusanagi & Yamashita, 2013; Tamura
& Kusanagi, 2014). That is, the effect of the task condition affects the performance, depending on the linguistic factors. Kusanagi and Yamashita (2013) and its partial replication study, Tamura and Kusanagi (2014), investigated this task–structure interaction by using grammaticality judgment tests with time constraints. In both studies, the effect of time pressure was observed in some of the target structures but not in the others, indicating that not only do task characteristics and linguistic factors impinge on L2 learners’ performance but also there is a clear interaction between task-related factors and linguistic structures.
One of the research areas that captures the clear differences between L1 and L2 sentence processing is online (in)sensitivity to number mismatches (e.g., Jiang, 2004,
2007; Jiang, Novokshanova, Masuda, & Wang, 2011). See the sentences in (1a) and (1b), where quantifier–noun number mismatches occur. A series of Jiang’s studies reported that L2 learners with some specific L1s (Chinese and Japanese) did not show an increase in reading times in the number mismatched sentences, as can be seen in (1a) and (1b), indicating that those learners might be insensitive to the number agreement.
(1a) She picked a few of her [*dress/dresses] and left quickly. (1b) The professor noticed a few of his [friend/*friends] in the picture. (Jiang, 2007)
Another interesting finding in Jiang’s studies was that the participants in these studies performed almost perfectly to detect the errors in offline grammatical judgment tasks, but were unable to do so in the online self-paced reading tasks. These contrasting findings led Jiang to claim that the L2 learners’ knowledge about number agreement was not fully integrated (Jiang, 2007). In other words, the L2 learners had only rudimentary implicit knowledge, or their skills remained unautomatized. Meanwhile, as Jiang (2004, 2007) reported, the L2 learners performed similarly to the native speakers of English when processing sentences measuring the knowledge of verb-subcategorization. The different performance among the structures that L2 learners exhibited was explained in terms of the selective integration of linguistic knowledge (Jiang, 2007).
1.2. Theoretical Framework of the Study
Regarding the acquisition of number features, linguistics-based second language acquisition researchers have offered numerous hypotheses such as (a) The Missing Surface Inflection Hypothesis (e.g., Prévost & White, 2000), (b) The Representation Deficit Hypothesis (e.g., Hawkins, 2005), (c) The Morphological Congruency Hypothesis (Jiang et al., 2011), (d) The Transfer Hypothesis (Nicol & Greth, 2003), and (e) The Interpretability Hypothesis (Tsimpli & Dimtrakopoulou, 2007). These hypotheses, which will be briefly reviewed later on, explain the reasons for difficulty in acquiring number features, processing number agreement, and attaining native-like processing skills.
The Missing Surface Inflection Hypothesis poses that L2 learners do have the abstract morphosyntactic features such as tense and grammatical number, but they are
not capable of mapping these features onto the surface morphemes (Prévost & White, 2000) due to cognitive overload. This can be called the computational difficulty position. This position is slightly different from that which argues that the problem with morphology in L2 acquisition lies in impairment of interlanguage grammar itself; that is, L2 learners struggle with morphological manipulation at the level of abstract features or categories.
From another point of view, the Representation Deficit Hypothesis and Interpretability Hypothesis can be called the first language influence position because they put more emphasis on the learners’ L1s; if a learner does not have a given formal feature (also called an uninterpretable one in the framework of the Interpretability Hypothesis) in his/her L1, the person cannot acquire the feature after the critical period (Hawkins, 2005; Tsimpli & Dimtrakopoulou, 2007, for instance).
Similarly, the Morphological Congruency Hypothesis proposed by Jiang et al. (2011) also looks at the role of L1 in L2 acquisition, though it focuses instead on the congruency of morphological systems between an L1 and an L2. Taking an instance in the case of sensitivity to number agreement, it poses that the existence of the morphological systems with subject–verb agreement or inflection for grammatical number in L1s can be used to predict one’s attainability of native-like processing skills. Nicol and Greth (2003) similarly claim that bilinguals (including L2 learners) tend to transfer their L1 strategies to L2 subject–verb agreement. This framework predicts that the speakers of L1s which do not have subject–verb agreement processes cannot transfer their processing strategy from the first language, and such miss-transfers and non- transfers result in non-native-like performance.
More generally, The Shallow Structure Hypothesis (SSH, Clahsen & Felser, 2006), based on their comprehensive review of L1 and L2 processing studies, proposed that L2 learners have a tendency to underutilize syntactic and morphosyntactic information during their online sentence processing, and to rely much more on lexical, semantic, and pragmatic information, unlike native speakers (Clahsen & Felser, 2006). The Good Enough Approach (e.g., Lim & Christianson, 2013), on the other hand, argues that even in the case of L1 sentence processing, comprehenders tend to rely much more on the guidance of lexical, semantic, and pragmatic (or maybe probabilistic) information, depending on the purpose of the processing (or simply the effects of task-related factors).
The present study was undertaken to provide empirical evidence for the inquiry from a relatively new angle. We focus on L2 learners’ sensitivity to notional numbers in subject–verb agreement processing, of which very little is reported in the literature of L2 studies.
1.3. Focus of the Present Study
In a traditional framework of studies on subject–verb agreement, there are three roles of grammatical constituents involved in subject–verb agreement. The first one is called the agreement controller. This controller is, in most cases, the head noun (subject) in a sentence. Predicates (main verbs) also serve in the role of agreement target, of which the number must be matched with the number of the agreement controller.
In the cases that a subject forms a complex noun phrase (NP) in which a local prepositional phrase (PP) is attached, the local noun has the role of an agreement attractor, which is fundamentally irrelevant to (target-like) number agreement, but sometimes affects even native speakers’ number agreement processes (this phenomenon is also called broken agreement, see Bock & Cutting, 1992; Bock & Eberhard, 1993; Jiang, 2004; Pearlmutter, Garnsey, & Bock, 1999). Specifically, it is well-known that mismatches of numbers between an agreement controller and an agreement attractor cause more agreement errors in production data and delay in reading time in native speakers’ online comprehension. The tree diagram in (2) represents the three roles involved in subject-verb agreement of numbers.
in Det Nattractor
likepl Det N
Interestingly, L2 learners were not affected by the effect of the numbers of local nouns (agreement attractors), unlike native speakers (Jiang, 2004). In other words, the grammatical attraction phenomenon was not evident in the case of online reading comprehension of L2 learners, at least with some specific L1 backgrounds such as Chinese and Japanese. Thus, it is supposed that L2 learners are prone to underutilize the grammatical number features; however, there remains a question of whether L2 learners make use of notional number features in processing the number agreements.
Some words, such as everyone, are, needless to say, grammatically singular, but notionally (conceptually or semantically) plural since they refer to multiple referents― people in the same group. This can be applied to almost all collective nouns such as family, team, and band. Although it is well known that subject–verb number agreement in English is purely syntax-driven, such a notional number of a local noun is sometimes considered to affect native speakers’ sentence processing (e.g., Bock & Cutting, 1993; Bock & Eberhard, 1993; Eberhard, 1999; Hoshino, Dussias, & Kroll, 2010; Humphrey & Bock, 2005). This can be called the notional number attraction phenomenon. However, very little is known about the phenomenon in the case of L2 learners.
The studies of Jiang (2004, 2007) and others (e.g., Lim & Christianson, 2014) on L2 learners’ online insensitivity to number mismatches lead us to anticipate that L2 learners are possibly independent of the notional number attraction phenomenon as well as the grammatical one. Meanwhile, SSH leads us to predict that L2 learners’ number matching processes will be affected by the notional number of the attractor adjacent to the verb and not by the grammatical one, since L2 learners’ online processing is supposed to be guided by lexical, semantic and pragmatic information.
Moreover, Inagaki’s (2014) study provided empirical evidence that Japanese EFL learners tended to rely on semantic information when judging mass-count distinction in English. Inagaki explained that this might be due to L1 influence; in other words, in Japanese, whether or not a noun is mass or count is determined semantically because Japanese does not have a grammatical marker representing number features. In English, however, it is the syntactic cue that is important for identifying mass-count distinctions. Because of the failure to map syntax to conceptual information due to L1 influence, Japanese learners have difficulty disambiguating the mass-count distinction.
2. Research Question
Given the assumptions on the basis of the previous research mentioned above, it is plausible to presuppose that L2 learners mistakenly utilize the notional number of nouns during their online sentence processing, while the grammatical number of nouns is considered not to be fully represented in L2 learners’ mental lexicons, or is inaccessible during online tasks.
Thus, the present study addressed the following research questions.
1. Are Japanese L2 learners affected by notional number attraction during online sentence processing?
2. Are Japanese L2 learners affected by grammatical number attraction during online sentence processing?
In order to answer the research questions above, the present study adopted the reaction time procedure. If an increase in reading times was observed when the Japanese learners of English read the sentences in which notionally plural corrective nouns function as attractors, this might be empirical evidence that Japanese L2 learners are not insensitive to all types of number agreement phenomena. It means that there is a possibility that Japanese learners of English might be thinking about the number features of nouns conceptually during their online reading processes. Besides, if the reading times did not differ in the case of grammatical number attraction conditions, it would be a striking contrast to the case of notional number attraction conditions.
3. The Present Study 3.1. Participants
The participants of the present study were composed of graduate and undergraduate university students who lived in Japan (N = 28). They usually learn and use English for academic purposes and it is their most fluent foreign language. The population we targeted to measure were highly proficient L1-Japanese learners (EFL learners), and all those in the sample were confirmed to be matched with the characteristics of the population using an interview and a questionnaire. The academic majors of the participants included international economics, linguistics, international culture, and
education. Table 1 summarizes the demographic information about the sample. Since almost half of the sample had previous experience of taking a computer-based self-paced reading task, it would reduce the noise which might happen during the experiment and skew the result of the study. All the participants received financial compensation.
The sample size of the present study was 28. The previous studies which adopted similar designs, such as Jiang (2004, 2007) also had the same size as the present study. Moreover, a priori power analysis was conducted in order to estimate the required sample size. The research design of the present study was expected to adopt a MANOVA followed by an ANOVA, and it is the ANOVA that requires a larger sample size which meets the estimated medium effect size, error probabilityα = .05, and targeted statistical power, 1-β = .80. Therefore, a priori simulation for ANOVA was conducted. The result showed that the total sample size which met all the criteria mentioned above was exactly 28, indicating that the sample size, 28, was not too small to detect the middle level of effect sizes.
Table 1. The Descriptive Statistics of the Participants’ Demographic Information
M SD Skewness Kurtosis
TOEIC 779.70 127.37 -0.17 -0.57
Age 24.14 3.88 1.47 3.18
Reading 3.14 1.04 -0.30 -0.86
Writing 3.23 0.92 -0.10 -1.14
Listening 3.05 1.05 0.18 -0.40
Speaking 2.91 0.92 -0.61 -0.15
Vocabulary 2.73 1.03 0.04 0.02
Grammar 3.05 1.17 -0.29 -0.47
Note. The self-report proficiency ratings can range from one to five. N = 28.
The self-paced reading task was conducted in the word-by-word and moving- window style. The program was created using Hot Soup Processer 3.2, and a 15 inch- screen, Windows laptop was used. In the very first step of the program, the positions for
the words were presented by black underlines. Then, once the right arrow key was pressed, the first word appeared. Each time the button was pressed, the word on the monitor disappeared and the next word appeared as described in Figure 1. The time to press the button for each word was measured by the program in milliseconds.
After reading each of the filler sentences, yes/no questions about the contents of the sentences were attached in order to confirm the comprehension levels of the participants (detailed in a later section).
In total, the present study created six types of stimuli as summarized in Table 2. There were three types of attraction conditions: (a) control (no attraction), (b) notional attraction, and (c) grammatical attraction. Each attraction type had two levels; (a) grammatical (number matched), and (b) ungrammatical (number mismatched).
In all the types of stimuli, we used only a single word, everyone, for the subjective head noun (the agreement controller) of the sentences, in order to observe the effect of attractions without any other types of effect caused by agreement controllers. The word everyone is grammatically singular, but notionally plural. The word quite often has PPs attached, such as everyone in the room, and everyone in the family, and these phrases are likely to be pragmatically plausible.
Figure 1. The self-paced reading task with the moving window version. The target sentence was presented word by word.
The nouns in attached PPs varied among the three attraction conditions. In the control condition, the local nouns were common nouns of which the grammatical and notional number were singular (e.g., room, office, restaurant, school, house, bar, park, and theater). In the notional attraction condition, collective nouns were used (e.g., team, company, audience, crew, family, class, and band). These words were grammatically singular, but notionally plural. In the grammatical attraction condition, inflected normal words such as stands, cars, films, pictures, and pages were used. These words became both grammatically and notionally plural.
The agreement targets also had two versions; matched and mismatched. We used only the past tense of the copula (was for matched, and were for mismatched) to reduce the variance of reading times caused by diversity of lexical items. In all of the conditions, since the agreement controller everyone was grammatically singular, the agreement targets had to be singular as well. The combinations of the conditions and their examples are summarized in Table 2.
Table 2. The Conditions and Examples of the Stimuli
Attraction Agreement Attractor Target Example
Control Matched [sg] [sg] [sg] Everyone in the room was
Mismatched [sg] [sg] [pl] Everyone in the room were
Notional Matched [sg] [pl] [sg] Everyone in the team was
Mismatched [sg] [pl] [pl] Everyone in the team were
Grammatical Matched [pl] [pl] [sg] Everyone in the halls was
Mismatched [pl] [pl] [pl] Everyone in the halls were Note. G = grammatical number, N = notional number.
Each condition had four items ranging from eight to ten words so that in total, 24 target stimuli were used in the present study. Additionally, in order to distract the
attention of the participants from the target grammatical phenomenon, the stimuli list included 24 other sentences as filler items, which were equal to exactly half the total number of items.
The comprehension questions were attached to 24 sentences out of 48. The mean accuracy score of the participants was .96, by which we judged that the participants had engaged well in the reading tasks, and they had been successfully made to focus on the contents of the stimuli.
In order to compare the reading times among the conditions, we set the five target regions, as can be seen in (3). The first region (region A) was set at the position of the determiner in the PPs. Region B was for the agreement attractor where the words varied. Region C, which was the main interest region in the present study, was located at the position of the agreement targets. Moreover, two more regions (D and E) were set in order to observe the delayed and spillover effects of reading times.
(3) Everyone |in |A the |B room |C was/were |D writing |E the …
To implement data cleaning on the reading times, we conducted outlier processing. We removed more than M ± 2SD reactions from the data. The removed data was almost 12% of the entire data. The data without outliers was still positively skewed and did not fit the normal distribution. Therefore, the log-transformation method was used so that further parametric analyses could be safely conducted. By using a transformation with the base of 10, the data followed the normal distribution.
In order to detect the reading time differences, we first conducted a MANOVA whose design was five regions × three types of attractions × two target versions (matched and mismatched). Then, ANOVAs for the data of the main interest and its next region (region C and D) were performed. In addition, in order to discuss the strength of the attraction effects more directly, the point and interval estimation method for the effect sizes (Cohen’s d, Hedge’s g, and r) was used.
The descriptive statistics for the row reading time data in the five regions are
summarized in Table 3. Figure 2 graphically represents the time courses of the participants’ reading times. Apparently, there were greater variances in reading times in the main interest region, which was probably due to the attraction effects. In addition, variances were also found in region D, where the delayed and spillover effects should emerge.
Table 3. The Descriptive Statics of the Reading Time Data
A B C D E
M SD M SD M SD M SD M SD
Control Matched 321 81 490 159 466 100 585 163 466 111
Mismatched 315 65 486 161 549 173 645 219 436 95
Notional Matched 334 86 511 210 569 182 732 244 448 88
Mismatched 338 77 513 188 558 183 699 187 453 102
Grammatical Matched 326 72 455 151 532 197 643 176 449 103
Mismatched 293 76 467 128 583 200 644 192 449 107
Note. N = 28.
Then, the transformed data were submitted to a MANOVA. The results based on the Figure 2. The line plot representing the time courses of the reading time.
The data were based on the raw reading times (ms). 800
Mean Reading Time (ms) Control Matched
Region A Region B Region C (Target)
Region D Region E Control Mismatched Notional Matched
Notional Mismatched Grammatical Mismatched
Pillai’s trace were that the main effects of the attractions on the five regions were statistically significant, F(10, 102) = 4.21, p < .01, ηp2 = .29, the main effects of grammaticality (matched or mismatched) did not reach statistical significance, F(5, 23) = 1.23, p = .33, ηp
2 = .21, and the interactions were also statistically insignificant, F(10, 102) = 0.25, p = .15, ηp
2 = .13. As the overall tendency, there was no statistically significant interaction. Then, we analyzed the effects in the main interest region (region C), and the next region (region D) to find the delayed and spillover effects.
The ANOVAs for the regions C and D were performed using the Greenhouse– Geisser’s adjustment because of the violations of the assumption of sphericity with some variables. The results in region C showed that the main effect of attractions was statistically significant, F(2, 54) = 3.81, p = .03,ηp2 = .12, the main effect of grammaticality was not statistically significant, F(1, 27) = 0.94, p = .34, ηp
2 = .03, and the interaction was statistically significant, F(2, 54) = 4.34, p = .02, ηp2 = .14. Since the interaction reached statistical significance, analyses of the simple main effects for the interaction were conducted. The results were that the simple main effect of grammaticality in the control condition was statistically significant, F(1, 27) = 5.49, p = .03, ηp2 = .17, the simple main effect of grammaticality in the notional attraction condition was not statistically significant, F(1, 27) = 0.69, p = .41, ηp2 = .02, and the simple main effect of grammaticality in the grammatical attraction condition was also not statistically significant, F(1, 27) = 0.06, p = .81, ηp
2 < .01. Subsequently, the simple main effect of attractions in the grammatical (matched) condition was statistically significant, F(1, 27) = 6.90, p < .01,ηp2 = .20, while the simple main effect of attractions in the ungrammatical (mismatched) condition was not statistically significant, F(10, 102) = 1.45, p = .24, ηp
2 = .01. Since the main effect of attractions was also statistically significant, multiple comparisons were performed using Holm’s adjustment of alpha levels. The results showed that only the difference between the control and grammatical attraction conditions was statistically significant, t(27) = 3.45, p < .01.
Next, the results of the ANOVA in region D were that the main effect of attractions was statistically significant, F(2, 54) = 9.88, p < .01, ηp2 = .27, the main effect of grammaticality did not reach statistical significance, F(1, 27) = 0.56, p = .46, ηp2 = .02, and the interaction was also not statistically significant, F(2, 45) = 1.33, p = .27, ηp
.04. Again, since the main effect of attractions was statistically significant, multiple
comparisons were performed in the same way as for region C. The results showed that the difference between the control and notional attraction conditions was statistically significant, t(27) = 4.17, p < .01, and this was also true between the notional and grammatical attractions, t(27) = 3.53, p < .01. Figures 3 and 4 graphically represent the results in region C and D respectively.
Furthermore, we estimated the effect sizes for each attraction. In region C, the effect sizes in the control condition showed a medium level, d = 0.48 [0.16, 0.79], g = 0.46 [0.16, 0.77], r = .53. The notional number attraction showed none to a small level, d = 0.07 [-0.33, 0.47], g = 0.07 [-0.32, 0.46], r = .07, and the grammatical attraction exhibited a small level, d = 0.25 [-0.09, 0.60], g = -0.09 [-0.02, 0.59], r = .28.
In region D, the control condition showed a small level, d = 0.48 [0.16, 0.79], g = 0.46 [0.16, 0.77], r = .53, while the notional and grammatical attraction conditions did not show a larger level of effect sizes. d = 0.08 [-0.21, 0.37], g = 0.08 [-0.20, 0.36], r = .11, d = 0.08 [-0.21, 0.37], g = 0.08 [-0.20, 0.36], r = .11, respectively. The scatter plots in Figure 5 show the correspondences of each case in the three attraction conditions in regions C and D.
Figure 3. Plot representing the result in region C.
Figure 4. Plot representing the result in region D.
Control Notional Attraction Condition
Log Transformed Mean Reading Times
Control Notional Attraction Condition
Log Transformed Mean Reading Times
5.1. Representation Vulnerability
The results indicated that both the notional and grammatical number attraction phenomena in L2 were found, showing the insignificant reading time differences between the notional and grammatical attraction conditions in the main interest region, and the next region. This means that the participants were affected by the effects of both the notional and grammatical numbers of local nouns within PPs (agreement attractors) on their real-time reading process.
As has been discussed in the background section, L2 learners were supposed to face difficulty in representing number features or accessing such morphosyntactic features in online tasks. Nevertheless, the participants in the present study obviously showed their online sensitivity to number mismatches in the control condition as the reading time of the singular copula was faster than that of the plural. Moreover, the results in the grammatical attraction condition showed insignificant reading times, which may suggest Figure 5. Scatter plots representing the correspondences of the mean reading times of individuals in each condition. The line with breaks represents y = x. The cases above the line indicate the participants who read the mismatched copula longer.
2.6 2.8 3.0 3.2 Control at C
2.6 2.8 3.0 3.2 Notional Attraction at C
2.6 2.8 3.0 3.2 Grammatical Attraction at C
2.6 2.8 3.0 3.2 Control at D
2.6 2.8 3.0 3.2 Notional Attraction at D
2.6 2.8 3.0 3.2 Grammatical Attraction at D
that the participants’ online sentence processing was attracted by the number features of local nouns, similarly to native speakers. This may indicate that the participants represented and fully accessed both of the grammatical number features of the controller (everyone) and the target (was), against the observations of several previous studies (e.g., Jiang, 2004, 2007; Jiang et al., 2011). Thus, the findings in the present study, unlike the previous studies, do not fully support the hypotheses regarding to the critical roles of L1s in the acquisition and the online processing of grammatical number (the L1 influence position). Rather, our hypothetical argument here is that L2 learners might basically be capable of representing and accessing the number features just as native speakers do. This assumption may support the computational difficulty position, such as the missing surface inflection hypothesis (e.g., Prévost and White, 2000).
However, it was also observed that the participants were guided by the attraction of notional numbers, and its effects were greater than the counterpart of grammatical attraction. This clearly supports the claims in SSH, which are that L2 learners’ processing relies more on lexical, semantic, and pragmatic information than syntactic and morphosyntacitic information. This semantic-based processing tendency that the learners demonstrated seems to align with the findings of Inagaki’s (2014) study. Inagaki pointed out that Japanese learners were likely to decide about the countability of nouns conceptually on account of L1 influence. From this point of view, it might be premature to conclude that there was no L1 influence in considering the mechanism of Japanese L2 learners’ sentence processing. The L1 influence is not ‘critical’ but it does affect the way learners deal with number agreement. What actually happens in L2 online sentence processing is much more complicated than is expected.
This contradictory interpretation might be resolved if we suppose that Japanese L2 learners’ number representation and its utilization were easily interchangeable between the grammatical and notional numbers, being influenced by a semantic-driven processing tendency which can be explained both by SSH or L1 influence. To put it differently, it is possible to infer that the participants mistakenly utilized the notional numbers instead of the grammatical ones that were solely responsible for number agreement in English. This observation led us to assume that L2 learners’ representation of number features is not deficient but simply vulnerable.
Our assumption on representation vulnerability values much more the realistic
nature of L2 learners’ processing than just L1 influences. It is an established fact that L2 learners have difficulty in attaining native-like grammatical performance. However, L2 learners exhibit a greater variety of performance than do native speakers, heavily depending on tasks, situations, purposes of language use, and grammatical structures in use. This may be because, at every moment, L2 learners flexibly and strategically compensate their vulnerable grammatical information or formal features with cues of notional, semantic, conceptual and contextual information in order to construct an appropriate situation model in their minds.
5.2. Limitations and Future Directions
Several limitations should be noted in interpreting the results of this study. First, it should be acknowledged that the result of the present study should not be generalized to L2 learners of English in general because the population of this study was not all L2 learners from different backgrounds. In addition, the sample of this study could not represent all Japanese L2 learners. Instead, this study focused on proficient Japanese learners of English with mean scores of TOEIC of around 700–800. Since it is plausible that learners’ proficiency might affect the way they process sentences online, it is necessary for future research to take into account the effect of English language proficiency1. Nonetheless, this does not mean that the sample size of the present study was too small, as ensured by the priori power analysis.
Second, as to the issue of generalizing the results of the present study, L1 influence, which was repeatedly used in this article, should be carefully interpreted. Since only Japanese learners of English participated in the study, the observed increase in reading time in the notional attraction condition might not be due to Japanese learners’ tendency to give number features to a noun semantically. Future research should investigate learners from different L1 backgrounds other than Japanese in order to make sure that conceptually-driven number agreement processing is peculiar to Japanese learners of English.
Third, it should be noted that in all the stimuli the agreement controller was everyone. In addition, only the past tense form of copula be was used. This kind of strict control of the stimuli was inevitable to remove various factors that might affect the way L2 learners process the sentence. However, it also came in for severe criticism because if
agreement controllers had been different nouns, the results might have been completely different. In addition, it is desirable to see various agreement patterns. The agreement controllers, attractors, and targets can all be either singular or plural; therefore, the possible total number of agreement patterns is eight, theoretically. In order to comprehensively reveal how L2 learners process number agreements, further studies which incorporate as many of the factors mentioned above as possible are needed.
Lastly, it should be pointed out that although the present study adopted the reading time procedure with self-paced reading tasks, it is necessary for future research to investigate the L2 learners’ performance in different types of task because task characteristics are one of the critical factors affecting learners’ performance, as Lim and Christianson (2014) pointed out.
Since the current study is exploratory in nature, a number of limitations discussed in the earlier section should be addressed in future research. Nonetheless, it can be said that the main contribution of the present study is to provide the first empirical evidence for the notional number attraction phenomenon in L2 learners’ online sentence processing. The result of the self-paced reading tasks provide suggestive data that Japanese L2 learners seem to have a tendency to process number features conceptually and semantically, and this results in making their number representation vulnerable. However, the results also indicate that Japanese L2 learners might not be totally insensitive to number agreement errors because an increase in reading time was, to some extent, observed in the control conditions and grammatical attraction conditions. Further investigation is necessary to get a clear picture of L2 learners’ number agreement processes by the use of various types of agreement and vocabulary with different L1 speakers.
1. One reviewer recommended providing some proficiency information of the participants in the previous studies such as Jiang’s (2004, 2007) studies in order to compare the proficiency of the both present and previous studies. The mean score of TOEFL scores reported in Jiang (2004), for example, was 608.1 in Experiment 1 and
604.4 in Experiment 2, and 618.56 in Jiang’s (2007) study. Although direct comparison of the TOEFL and TOEIC scores cannot be made, it was considered that the participants of Jiang’s (2004) study were much more proficient than the participants in the present study.
Bock, K., & Cutting, J. C. (1992). Regulating mental energy: Performance units in language production. Journal of Memory and Language, 31, 99–127.
Bock, K., & Eberhard, K. M. (1993). Meaning, sound and syntax in English number agreement. Language and Cognitive Processes, 8, 57–99.
Bock, K., & Miller, C. A. (1991). Broken agreement. Cognitive Psychology, 23, 45–93. Clahsen, H., & Felser, C. (2006). Grammatical processing in language learners. Applied
Psycholinguistics, 27, 3–42.
Eberhard, K. M. (1999). The accessibility of conceptual number to the processes of subject–verb agreement in English. Journal of Memory and Language, 41, 560–578. Hawkins, R. (2005). Explaining full and partial success in the acquisition of second
language grammatical properties. Second Language, 4, 7–25.
Hoshino, N., Dussias, P. E., & Kroll, J. F. (2010). Processing subject–verb agreement in a second language depends on proficiency. Bilingualism: Language and Cognition, 13, 87–98.
Humphrey, K. R., & Bock, K. (2005). Notional number agreement in English. Psychonomic Bulletin and Review, 12, 689–695.
Inagaki, S. (2014). Syntax–semantics mappings as a source of difficulty in Japanese speakers’ acquisition of the mass-count distinction in English. Bilingualism: Language and Cognition, 17, 464–477.
Jiang, N. (2004). Morphological insensitivity in second language processing. Applied Psycholinguistics, 25, 603–634.
Jiang, N. (2007). Selective integration of linguistic knowledge in adult second language learning. Language Learning, 57, 1–33.
Jiang, N., Novokshanova, E., Masuda, K., & Wang, X. (2011). Morphological congruency and the acquisition of L2 morphemes. Language Learning, 61, 940– 967.
Kusanagi, K., & Yamashita, J. (2013). Influences of linguistic factors on the acquisition of explicit and implicit knowledge: Focusing on agreement type and morphosyntactic regularity in English plural morphemes. Annual Review of English Language Education in Japan, 24, 205–220.
Lim, J. H., & Christianson, K. (2013). Integrating meaning and structure in L1–L2 and L2–L1 translations. Second Language Research, 29, 233–256.
Lim, J. H., & Christianson, K. (2014). Second language sensitivity to agreement errors: Evidence from eye movements during comprehension and translation. Applied Psycholinguistics, Advanced online publication. doi:10.1017/S0142716414000290 Nicol , J., & Greth, D. (2003). Production of subject–verb agreement in Spanish as a
second language. Experimental Psychology, 50, 196–203.
Pearlmutter, N. J., Garnsey, S. M., & Bock, K. (1999). Agreement processes in sentence comprehension. Journal of Memory and Language, 41, 427–456.
Prévost, P., & White, L. (2000). Missing surface inflection or impairment in second language acquisition? Evidence from tense and agreement. Second Language Research, 16, 103–133.
Roberts, L., & Meyer, A. (2012). Individual differences in second language learning: Introduction. Language Learning, 62, 1–4.
Tamura, Y., & Kusanagi, K. (2014). Countability of normal/material nouns in Japanese EFL learners’ explicit and implicit knowledge. Paper presented at 40th Annual Conference of Japan Society of English Language Education, Tokushima, Japan. Tsimpli, I. M., & Dimitrakopoulou, M. (2007). The Interpretability Hypothesis:
Evidence from wh-interrogatives in second language acquisition. Second Language Research, 23, 215–242.