Age of Arrival (AOA) effects on anaphor processing by Japanese bilinguals

1 Introduction

There has been a great deal of research on bilingual speakers’ linguistic performance in their first language (L1) and second language (L2). Ample evidence shows that bilinguals’ L1 knowledge affects their L2 performance. However, it has also been shown that bilinguals’ L1 performance can be affected by their L2 knowledge. As an extreme case of L2 influence, several studies observed that one’s L1 performance is attritted by L2 exposure. For example, De Bot et al. (1991) reported that Dutch immigrants living in France and having limited exposure to their native Dutch showed a decrease of their Dutch proficiency. Also, Carson and Kuehn (1994) found that native Chinese speakers living in the U.S. experienced a decrease in their L1 writing skills as their L2 writing skills increased. It is this latter type of influence which is the focus of the current paper.

In the present paper, we re-examine the data and results reported in Shoji et al.’s (2016a) study of Japanese anaphor processing by Japanese speakers residing in the U.S. Because the L2-immersed bilingual Japanese speakers tested in Shoji et al. represent a population whose L1 processing is liable to be influenced by L2, it is a good study to subject to an examination of whether and how much exposure to L2 can affect L1 performance. This re-examination was motivated by another recently completed study (Shoji et al. 2016b) which tested functionally monolingual Japanese speakers residing in Japan, and which had different results from the earlier (Shoji et al. 2016a) study.^[1] That the results of these two studies differ despite the very similar experimental design and materials, led us to consider the possibility that L2-immersed bilingual participants exhibited L1 attrition in their performance in the task. Thus, in the present paper, we compare the results from Shoji et al. (2016a, 2016b) using one measure of exposure to L2, Age of Arrival (AOA) to the US, as the dependent variable.

2 L2-immersed bilinguals versus L1-immersed monolinguals

3 Age of Arrival (AOA)

Age of Arrival (AOA) is the age at which L2 speakers arrive at an area where their L2 is regularly spoken and from which they are regularly exposed to their L2. A number of studies on bilinguals have observed AOA effects on their L1 performance. These studies suggest that the earlier a bilingual speaker is exposed to an L2 environment, the more probable he/she will lose aspects of their L1. For example, a study by Hakuta and D’Andrea (1992) on U.S. residents with Mexican background showed that the participants whose AOA was under 10 years of age failed to maintain native-like L1 ability, relative to those whose AOA was greater than10 years. Also, Yukawa’s (1997) longitudinal research on bilingual Japanese prepubescent children reported that the younger their AOA, the more likely that they would suffer L1 attrition. Ammerlaan’s (1996) study on Dutch immigrants in Australia (AOA: 6 to 29) as well as Pelc’s (2001) study on Greek immigrants in the U.S. (AOA: 8 to 32) observed similar effects: late AOA favors L1 maintenance. If Shoji et al.’s (2016a) bilingual Japanese speakers are also affected in a similar manner, then relatively young AOA participants would likely suffer relatively greater L1 attrition as compared with relatively older AOA participants.

The average AOA of all participants in Shoji et al. (2016a) was 27.29 years old, with a range of age from 13 to 47 (SD=8.26). In our re-examination of the data, we first included participants’ AOA as a fixed factor in addition to antecedent and anaphor in a mixed effects model analysis in order to see if AOA interacts with antecedent and anaphor. In this reanalysis, we followed the same procedures for data preparation and analysis as reported in Shoji et al. (2016a), except that we added AOA as a fixed factor to the model. Our analysis detected an interaction between antecedent, anaphor and AOA (χ²(13)=27.703, p=0.010). The parameter estimates of the chosen model are shown in Table 2.

The table above exhibits significant or marginally significant interactions between AOA, antecedent and different types of anaphors (except overt-pronoun-ga), revealing that participants’ AOAs indeed affected anaphor processing by participants in Shoji et al.’s (2016a) study.

3.1 Early AOA versus late AOA

In order to further examine how AOA affects anaphor processing by Shoji et al.’s (2016a) bilingual participants, we split participants into two subgroups based on the median value of their AOA: 36 participants whose AOA was below the median (25 years) and 36 participants whose AOA was above it. The average AOA of the early AOA group was 20.44 years old (ranging from 13 to 25, SD=2.60), and the average AOA of the late AOA group was 34.14 years old (ranging from 26 to 47, SD=5.94). Separately, for each group, we conducted analyses identical to those in Shoji et al. (2016a). Assuming, as we did, that early AOA correlates with greater L1 attrition, we predicted that the early (younger) AOA group would yield results that were more distinct from the results gathered in Shoji et al.’s (2016b) study of monolingual Japanese residents. In contrast, we predicted that responses by the late (older) AOA group would be relatively similar to the monolinguals. In other words, these three groups’ results were predicted to have a linear pattern: monolinguals sensitive to L1 morphemes → late AOA moderately sensitive to L1 morphemes → early AOA insensitive to L1 morphemes.

For the data analyses, data from participants whose accuracy rates were below 80 % were removed. This included four participants in the earlier AOA group and one participant in the later AOA group. The reading time data from trials in which participants answered the comprehension questions incorrectly were also removed, affecting 12.55 % of the data from the early AOA group, and 9.57 % from the late AOA group. Overall outlier removal (extreme reading times and those identified by an adjusted boxplot rule) affected an additional 5.06 % of the early AOA group’s data and 4.58 % of the late AOA group’s data. All other aspects of the analyses were the same as the preceding. The reading times of each type of anaphors for the early AOA group and late AOA group are shown in Figure 1.^[7]

Figure 1:

Mean reading times of second sentence with anaphor (a) early AOA group (b) late AOA group.

3.1.1 Early AOA group

Our omnibus analyses of the entire data set from the early AOA group did not detect any significant interaction between antecedent (subject, object) and anaphor (repeated-name-ga, repeated-name-wa, overt-pronoun-ga, overt-pronoun-wa, null pronoun), χ²(4)=7.394, p=0.117. We also conducted a series of separate analyses for each full reference form, following Shoji et al. (2016a). The analyses contrasted repeated-name-ga vs. null pronouns (χ²(1)=2.903, p=0.088), repeated-name-wa vs. null pronouns (χ²(1)=1.323, p=0.250), overt-pronoun-ga vs. null pronouns (χ²(1)=5.630, p=0.018), and overt-pronoun-wa vs. null pronouns (χ²(1)=3.364, p=0.067). The interaction term examined in all the analyses for the early AOA group are shown in Table 3.

Table 3:

Anaphor-antecedent interaction in early AOA group analyses.

	β	SE	t	p
Omnibus Analysis
Antecedent: Object	0.148	0.014	3.603	<0.001 **
Anaphor: Repeated-name-ga	0.176	0.041	4.278	<0.001**
Anaphor: Repeated-name-wa	0.103	0.040	2.581	=0.010**
Anaphor: Overt-pronoun-ga	0.241	0.041	5.863	<0.001**
Anaphor: Overt-pronoun-wa	0.179	0.041	4.345	<0.001**
Antecedent: Object x Anaphor: Repeated-name-ga	−0.102	0.058	−1.759	=0.079.
Antecedent: Object x Anaphor: Repeated-name-wa	−0.051	0.058	−0.882	=0.378
Antecedent: Object x Anaphor: Overt-pronoun-ga	−0.141	0.058	−2.419	=0.016*
Antecedent: Object x Anaphor: Overt-pronoun-wa	−0.114	0.058	−1.960	=0.050.
Separate Analysis: RNP – Null pronoun vs. Repeated-name-ga
Antecedent: Object	0.141	0.041	3.463	<0.001**
Anaphor: Repeated-name-ga	0.184	0.041	4.505	<0.001**
Antecedent: Object x Anaphor: Repeated-name-ga	−0.098	0.057	−1.706	=0.089.
Separate Analysis: RNP – Null pronoun vs. Repeated-name-wa
Antecedent: Object	0.153	0.041	3.772	<0.001**
Anaphor: Repeated-name-wa	0.120	0.040	3.011	=0.003**
Antecedent: Object x Anaphor: Repeated-name-wa	−0.066	0.057	−1.151	=0.250
Separate Analysis: OPP – Null pronoun vs. Overt-pronoun-ga
Antecedent: Object	0.148	0.043	3.447	<0.001**
Anaphor: Overt-pronoun-ga	0.253	0.043	5.860	<0.001**
Antecedent: Object x Anaphor: Overt-pronoun-ga	−0.145	0.061	−2.379	=0.018*
Separate Analysis: OPP – Null pronoun vs. Overt-pronoun-wa
Antecedent: Object	0.141	0.044	3.256	=0.001**
Anaphor: Overt-pronoun-wa	0.179	0.044	4.097	<0.001**
Antecedent: Object x Anaphor: Overt-pronoun-wa	−0.113	0.061	−1.839	=0.066.

1. Notes: Factors were coded with dummy coding. The null pronouns and subject antecedent were used as the reference levels (value=0) for the anaphor form and antecedent factors respectively. Reading times were log transformed and the coefficients thus represent changes between conditions in log scale. Significant effects at a p≤0.01 levels are marked with a **, at a p≤0.05 levels with a *, and at a p≤0.1 levels with a ‘.’.

As shown in the table above, the analyses found a marginally significant RNP with repeated-name-ga, but no RNP with repeated-name-wa. Also, a significant OPP was detected with overt-pronoun-ga but an only marginally significant OPP was found with overt-pronoun-wa. We noted that the RNP effect was weaker with wa-marked repeated names (topic) than with ga-marked repeated names (non-topic). Similarly, the OPP effect was found to be weaker with wa-marked overt pronouns (topic) relative to ga-marked overt pronouns (non-topic). These outcomes are similar to what L1-immersed monolingual participants in Shoji et al. (2016b) exhibited, i. e., the monolingual participants showed the RNP with ga-marked repeated names but not with wa-marked repeated names; they showed the OPP with ga-marked overt pronouns but not with wa-marked overt pronouns.

3.1.2 Late AOA group

The omnibus analysis of the data from the late AOA group detected a significant interaction between anaphor and antecedent: χ²(4)=11.011, p=0.026. Just like the analyses for the early AOA group, we conducted a series of analyses for each full reference form separately. The analyses contrasted repeated-name-ga vs. null pronouns (χ²(1)=3.544, p=0.060), repeated-name-wa vs. null pronouns (χ²(1)=7.755, p=0.005), overt-pronoun-ga vs. null pronouns (χ²(1)=5.692, p=0.017), and overt-pronoun-wa vs. null pronouns (χ²(1)=6.340, p=0.012). The results of all the analyses for the late AOA group are shown in Table 4.

Table 4:

Anaphor-antecedent interaction in late AOA group analyses.

	β	SE	t	p
Omnibus Analysis
Antecedent: Object	0.171	0.036	4.789	<0.001**
Anaphor: Repeated-name-ga	0.254	0.036	7.117	<0.001**
Anaphor: Repeated-name-wa	0.195	0.035	5.533	<0.001**
Anaphor: Overt-pronoun-ga	0.269	0.036	7.525	<0.001**
Anaphor: Overt-pronoun-wa	0.232	0.036	6.413	<0.001**
Antecedent: Object x Anaphor: Repeated-name-ga	−0.105	0.051	−2.075	=0.038*
Antecedent: Object x Anaphor: Repeated-name-wa	−0.150	0.050	−2.999	=0.003**
Antecedent: Object x Anaphor: Overt-pronoun-ga	−0.129	0.051	−2.530	=0.012*
Antecedent: Object x Anaphor: Overt-pronoun-wa	−0.128	0.051	−2.498	=0.013*
Separate Analysis: RNP – Null pronoun vs. Repeated-name-ga
Antecedent: Object	0.163	0.037	4.351	<0.001**
Anaphor: Repeated-name-ga	0.252	0.037	6.748	<0.001**
Antecedent: Object x Anaphor: Repeated-name-ga	−0.099	0.053	−1.885	=0.059.
Separate Analysis: RNP – Null pronoun vs. Repeated-name-wa
Antecedent: Object	0.169	0.037	4.556	<0.001**
Anaphor: Repeated-name-wa	0.196	0.037	5.334	<0.001**
Antecedent: Object x Anaphor: Repeated-name-wa	−0.145	0.052	−2.794	=0.005**
Separate Analysis: OPP – Null pronoun vs. Overt-pronoun-ga
Antecedent: Object	0.171	0.037	4.671	<0.001**
Anaphor: Overt-pronoun-ga	0.264	0.037	7.216	<0.001**
Antecedent: Object x Anaphor: Overt-pronoun-ga	−0.124	0.052	−2.394	=0.017*
Separate Analysis: OPP – Null pronoun vs. Overt-pronoun-wa
Antecedent: Object	0.171	0.036	4.730	<0.001**
Anaphor: Overt-pronoun-wa	0.240	0.037	6.515	<0.001**
Antecedent: Object x Anaphor: Overt-pronoun-wa	−0.130	0.052	−2.525	=0.012*

1. Notes: Factors were coded with dummy coding. The null pronouns and subject antecedent were used as the reference levels (value=0) for the anaphor form and antecedent factors respectively. Reading times were log transformed and the coefficients thus represent changes between conditions in log scale. Significant effects at a p≤0.01 levels are marked with a **, at a p≤0.05 levels with a *, and at a p≤0.1 levels with a ‘.’.

The table above shows that the late AOA group participants showed a marginally significant RNP with repeated-name-ga and a significant RNP with repeated-name-wa. Also, they showed significant OPPs with both overt-pronoun-ga and with overt-pronoun-wa. Overall, the late AOA group appears unaffected by different morphological markings (i. e., topic-wa or non-topic-ga), showing both processing penalties regardless of the morphemes. These outcomes are different from those observed with the L1-immersed monolingual participants in Shoji et al. (2016b).

Surprisingly, it was the early AOA group that performed similarly to the monolingual participants and not the late AOA group. The summary of the results from L1-immersed monolingual participants, the late-AOA bilingual group, and the early AOA bilingual group is shown below in Table 5 (cf. Table 1), which exhibits a non-linear pattern: monolinguals sensitive to L1 morphemes → late AOA insensitive to L1 morphemes → early AOA moderately sensitive to L1 morphemes.

Table 5:

RNP and OPP with each anaphor form for L-1 immersed monolinguals, L-2 immersed Late AOA, and L-2 immersed Early AOA.

Anaphors	Monolingual	Bilingual
		Late AOA	Early AOA
Null-pronoun vs. Repeated-name-ga	RNP	marginal RNP	marginal RNP
Null-pronoun vs. Repeated-name-wa	no RNP	RNP	no RNP
Null-pronoun vs. Overt-pronoun-ga	OPP	OPP	OPP
Null-pronoun vs. Overt-pronoun-wa	no OPP	OPP	marginal OPP

The table shows that monolinguals and the early AOA group were aware that wa-marking is different from ga-marking, but the late AOA group were insensitive to the different morphemes.

4 Discussion

According to our results, in the case of L2-immersed adults, the earlier their initial age of immersion is, the more likely they are to maintain sensitivity to L1-specific linguistic properties such as wa-marking in Japanese. In contrast, when an adult is immersed in their L2 relatively late, they tend to lose, or become insensitive to, L1-specific linguistic properties. These results actually contradict the prediction that we made based on the earlier studies, i. e., early AOA bilinguals are more likely to exhibit L1 attrition (Hakuta and D’Andrea 1992; Yukawa 1997; Ammerlaan 1996; Pelc 2001). However, notice that, as Köpke and Schmid (2004) point out, these earlier studies all included prepubescent AOA participants (i. e., none of them used only postpubescent AOAs). In contrast, Shoji et al.’s (2016a) participants were all postpubescent AOA bilinguals. Bylund (2009) argues that the nature and magnitude of L1 attrition might differ between pre- and postpubescent AOA bilinguals, and the present study appears to confirm this, suggesting that postpubescent AOA bilinguals exhibit an attrition pattern which is the opposite of prepubescent AOAs.

A possible explanation for these distinct attrition patterns among postpubescent bilinguals is provided by the Fundamental Difference Hypothesis (FDH) proposed by Bley-Vroman (1990), according to which different AOAs are associated with different liguistic systems for L1 and L2. While the FDH was proposed to explain how prepubescents and postpubescents differ in their construction of an L2 system, we would extend the FDH to explain different L2 systems in younger vs. older post-pubescent adults. This is because the onset of puberty itself, as it pertains to language learning, is likely not a trigger of abrupt change, but may instead reflect gradual changes associated with maturation across a wide range of ages (Lenneberg 1967; Bialystok and Hakuta 1994). Thus, we would extend the FDH to our explanation for how adult bilinguals differ among themselves. That is, among adult bilinguals, early AOA bilinguals have more access to universal linguistic principles and set parameters that are specific to different languages (the Principles and Parameters Theory, Chomsky 1980), whereas late AOA bilinguals who have less access to the universal principles tend to create a single unified language system via extensions either to their L1 system or to their L2 system.^[8] Thus, if their unified system is an extension of their L1, they will not incorporate a L2 property when it contradicts an analogous L1 property. Alternatively and conversely, if their unified system is based on their L2, then they fail to retain L1-specific linguistic properties when they clash with the analogous L2 properties. This creation of one unified language system is different from language-specific systems (i. e., independent parameter settings) that early AOA bilinguals build.

In the above explanation, we do not mean that a bilingual’s linguistic system is solely based on either L1 or L2. It is very possible that some aspects of the linguistic system are based on L1, while some others are based on L2. For instance, topic-marking with wa could be an L1 feature that our late AOA bilinguals did not retain because their interpretation of grammatical subjects was based on L2 English, wherein topic-subjects and non-topic-subjects are not differentiated. As a result, they did not show a different processing pattern between wa- and ga-marked anaphors, resulting in the RNP and OPP effects regardless of the morphological markings. On the other hand, results from the early AOA group showed their sensitivity to the morphological markings, as the antecedent-anaphor interactions were weaker when anaphors were marked by wa than when they were marked by ga. In other words, the results indicate that the early AOA group members retained their sensitivity to the L1-specific topic-marking, possibly because they had different parameter settings for L1 Japanese and L2 English.

A remaining question is why, as previous studies have shown, prepubescent young AOA bilinguals show drastic L1 attrition. They should have nearly full access to the universal linguistic principles and should also be able to set different parameter settings, as evidenced by the fact that children are more likely to be successful bilingual speakers than adult L2 learners (Bley-Vroman 1990). This fact could be explained by immaturity of the L1 system and extralinguistic matters, such as non-L1 contact and/or little L1 motivation. The early AOA children might only have a immature L1 system, one which can more easily be attrited or lost (e. g., Ventureyra et al. 2004), unlike the more settled L1 system of adult speakers. In addition, L1 attrition among younger bilinguals may also be attributable to the absence of motivation to preserve their L1 when they are in an environment in which their communication is mostly carried on in the L2 of their surroundings; it is well established that many children in this situation are not motivated to keep using their L1. Children who somehow maintain their motivation to retain their L1, as well as their contact with, and use of, it, tend to become more successful bilingual speakers, while those having neither L1 contact nor motivation to use their L1 will experience significant L1 attrition. Therefore, it is plausible that L1 attrition on the part of younger children is greatly attributable to environmental or psychological factors, whereas L1 attrition among adults (especially late AOA adults) is more likely due to aptitudinal factors.

5 Other factors: Length of residence (LOR) and age

In addition to AOA, we also investigated LOR and age effects for Shoji et al.’s (2016a) participants. LOR is the length of bilingual speakers’ residence in the environment where their L2 is regularly spoken.^[9]De Bot et al. (1991) observed that the longer the LOR is, the more likely bilinguals experience L1 attrition. The average LOR of all participants in Shoji et al. (2016a) was 5.841 years, ranging from 1 month to 24 years (SD=6.45). Also, the ages of participants reported in Shoji et al. (2016a) was 33.58 in average, ranging from 19 to 52 (SD=10.363). We ran analyses for LOR and age factors in the same manner as we did for AOA. When we included LOR or age as a fixed factor in addition to antecedent and anaphor in our analyses, we found a significant interaction between these factors, (LOR, χ²(13)=22.852, p=0.043; age, χ²(13)=23.782, p=0.033), indicating that both LOR and age indeed affected the participants’ anaphor-processing.

However, when we median-split the participants into subgroups based on the LOR or age, i. e., shorter LOR group vs. longer LOR group, and younger age group vs. older age group, respectively, our analyses did not generate any explanatory results for the comparison of Shoji et al.’s two studies (2016a, 2016b). Regarding LOR, the short LOR group did not show antecedent-anaphor interaction in its omnibus analysis (χ²(4)=6.124, p=0.190), and showed some interactions when separately analyzed for each anaphor: repeated-name-ga (χ²(1)=2.853, p=0.091), repeated-name-wa (χ²(1)=3.345, p=0.067), overt-pronoun-ga (χ²(1)=2.615, p=0.104), and overt-pronoun-wa (χ²(1)=5.247, p=0.022). On the other hand, the longer LOR group showed an antecedent-anaphor interaction in the omnibus analysis (χ²(4)=11.414, p=0.022), and showed significant interactions in separate analyses for each anaphor: repeated-name-ga (χ²(1)=4.290, p=0.038), repeated-name-wa (χ²(1)=4.516, p=0.033), overt-pronoun-ga (χ²(1)=10.15, p=0.001), and overt-pronoun-wa (χ²(1)=4.485, p=0.034).^[10]

As for the age factor, younger age group showed an antecedent-anaphor interaction in the omnibus analysis (χ²(4)=9.760, p=0.045), and showed some significant interactions when analyzed separately for each anaphor: repeated-name-ga (χ²(1)=2.444, p=0.118), repeated-name-wa (χ²(1)=7.247, p=0.007), overt-pronoun-ga (χ²(1)=6.258, p=0.012), and overt-pronoun-wa (χ²(1)=4.707, p=0.030). The older age group showed a marginal antecedent-anaphor interaction in the omnibus analysis (χ²(4)=8.798, p=0.066), and showed some significant interactions when analyzed separately for each anaphor: repeated-name-ga (χ²(1)=4.271, p=0.039), repeated-name-wa (χ²(1)=2.551, p=0.110), overt-pronoun-ga (χ²(1)=6.017, p=0.014), and overt-pronoun-wa (χ²(1)=5.951, p=0.012).

Importantly, none of these results in any group (regarding both LOR and age) exhibit the effects from different morphological markings. Therefore, we conclude that neither LOR nor age factor contributed to the different processing patterns that Shoji et al. (2016a, 2016b) exhibited for wa vs. ga.

6 Conclusion

This paper re-examined Shoji et al.’s (2016a) study by adding a new perspective that considered the L2 influence on participants’ L1 performance. In particular, our study revealed the different processing patterns depending on their AOA. This study thus contributes to the further understanding of language variability as affected by differential exposure to a second language and the initial age of that exposure.

Possible future research related to the present study could be the examination of LOR effect. Our analyses for LOR of Shoji et al.’s (2016a) participants show that, while the shorter LOR group tend not to elicit RNP or OPP, the longer LOR group showed the RNP and OPP for all types of anaphors. This might imply that the shorter LOR bilinguals tend to regard grammatical subjects as all topic-subjects whereas the late LOR bilinguals tend to regard them as all non-topics. This possibility would be an interesting issue to explore future research.