5.2 Results and discussion by task
5.2.3 The grammatical intuition data vs. translational production data:
5.2.3.2 Discussion of results divergence/variability across the two tasks
The discussion in the previous subsection made it evident that participants in proficiency subgroups, regardless of their L1 backgrounds, did not perform consis-tently across different task types. Even though their performance generally increased as their proficiency increased, the GJ task seemed to be much more difficult for L2 learners of English compared to the translation task, which appeared to be quite easy as far as embedded subjects are concerned. This contrast emerged when the results of the GJ task were compared to the results of the production task and disconfirm hypothesis H2 which predicted that all the learners, within each subgroup, would perform consistently across the two task types.
Generally speaking, the contrast within the performances of the learners appears consistent with the findings of previous empirical studies when considering the differences in the rates reported between subject omissions and null subject acceptances.124 These studies indicate that [+prodrop] learners acquiring a [-prodrop]
language consistently accept ungrammatical sentences with referential null subjects with approximate rates of 24–41% (Davies, 1996; Judy, 2011; Judy and Rothman, 2010; Orfitelli and Grüter, 2013; White, 1985, 1986), whereas such learners are reported to drop subjects when tested with production tasks only at the earliest stages of L2 development, if it exists, at approximate rates of 0–13% (Hilles, 1996;
Phinney, 1987; Orfitelli and Grüter, 2013).
The question that logically follows is this: What are the possible causes of such inconsistent use of subject pronouns (overt vs. null) noticed in the participants’
performance across both tasks (the intuitive and the translation tasks)? In other words, why do the learners persistently accept referential null subjects in the GJ task beyond the stage of L2 development when they have established the requirement for overt subjects in their production?
124 Notably, the results of several studies conducted to investigate the L2 learning of a wide variety of linguistic structures lend support to the observation that the use of different elicitation tasks would yield differences in the performance of L2 learners (Lee, 2014;
Tarone, 1985; Wright, 2010)
A possible explanation to account for this sort of inconsistency is to argue that [+prodrop] is part of the learners’ IL grammars of English, so that their ILs permit both null and overt referential subjects (this will be explained below more clearly when considering the notion of judgment strength of preferences). However, the written nature of the translation task prompted the learners to strongly prefer using overt pronouns over null forms, but when they are forced to make judgements about given sentences, null subjects may surface in their acceptability due to the fact that acceptability is a gradient concept.125
To investigate this argument in some depth, the participants’ strength of preferences for the overt pronouns over null forms need to be measured. One way to do so here (and recall what was discussed in relation to the rating scale used in the GJ task, how it worked, and how the data obtained were analysed in Chapter 4) is by looking at the results of the doubtful category (possibly incorrect option) only when the right corrections were provided; namely, only the PIT scores (refer to subsection 4.3.6.1: The GJ Task: The Adopted Marking Method). Figure 5:8 below displays the percentages of the PIT scores of the total rejected ungrammatical sentences with null subjects, which are made of the PIT scores added to the CIT scores (clearly incorrect—the true correction was provided):126
125 See section 4.3.2.1.2 for more detail about the differences between grammaticality and acceptability.
126 For more details, see to subsection 4.3.6.1: The GJ Task: The Adopted Marking Method.
Figure 5-8. The subgroups’ overall percentages of the PIT scores to ungrammatical sentences with null subjects.
The above figure illustrates the participants’ strength of preferences for overt pronouns over null forms; it shows that the French lower-intermediate subgroup participants, for example, by choosing the possibly incorrect option were reluctant/hesitant concerning 32.17% of the ungrammatical items they rejected. They have an inkling that these items could be correct, even though they corrected the errors that rendered the sentences ungrammatical. Similar observations can be noticed in all other subgroups of participants, regardless of the amount/size of the percentage of PIT rejection and/or the degree of doubt (little or great) when considering Figure 5.8 again.127 Hence, these PIT percentages can be taken as possible indices of the assumptions:
that choosing the possibly incorrect option indicates that the null subject option [+ null subject] is part of the learners’ IL grammar;
that managing to provide the right correction indicates a preference for the
127 Due to space limitations, only the results of the PIT scores of the French lower-intermediate subgroup is presented here as an example to illustrate and support the argument.
overt pronouns over the null ones; and
that choosing the possibly incorrect option and then managing to correct the sentence accurately support the idea that acceptability is a gradient concept.
Different explanations have been proposed to account for such intraindividual variability observed in the learner’s performance on the same day across any set of two or more tasks. For example, generative linguists have explained such variability by the fact that performance is not a perfect reflection of competence on all occasions. As they are interested in “what a speaker knows about language as an internal property of human mind rather than something external [the produced utterances]” (Chomsky, 1988, p. 36), they often argue for the use of GJ tasks to ascertain learners’ intuitions rather than using other tests that might not reflect the learner’s IL (see the relevant dissection in Chapter 2 and 4). Psycholinguists (e.g., Orfitelli and Grüter, 2013; Wright, 2010) seem to be more interested in learner’s variability even though they share the view with generative linguists that IL is a static/constant system at any given point in time; they link such variability to factors internal to the learner such as attention, processing limitation, and demands of short-term memory. In contrast, sociolinguists generally view IL as a dynamic linguistic system varying at any given point according to learner’s interaction with the environment (i.e., formal vs. informal context, speaking vs. writing task), which results in noticed variability (for more detail cf. Lowie and Verspoor, 2015; Throne, 1988;
Verspoor, Lowie, and van Dijk, 2008). Before leaving this discussion, it is important to comment briefly on these alternative explanations:
i. Various criteria have been implemented in the present study to make learner’s performance on both tasks reflect his/her IL to a great extent (see Chapter 4).
ii. If variability is only caused by extragrammatical factors, null subjects observed in L2 learners’ acceptances or productions must also have extragrammatical causes, not due to underlying grammars that permits them [+prodrop]. However, the findings of the present study and previous studies speak against this view; they show evidence of transfer of null subjects from
L1 to L2. I will come back to this point in more detail when the issue of null subject licensing is discussed in section 5.4.
iii. As for the sociolinguistic view of variability, I think it is impossible for tasks to alter the status of the IL; otherwise, a learner would be identified as being at two or more different developmental levels at the same time.
At this point, I turn to the notion of parameter resetting. To recap, the empirical data presented and discussed here provide evidence that while Finnish and Arabic L2 learners of English continued to accept null subjects even at the advanced stage of L2 development (despite the fact null subjects had disappeared from their speech during the early stages of acquisition), L1 French speakers fully converged on the English native-like usage of overt subjects. These findings are open to two quite different scenarios with respect to the null subject parameter resetting in L2A: the parameter either can be reset or cannot be reset.
According to the first scenario, one can argue based on the results that because the null subject would not be fully eliminated from the learners’ IL grammars—though it will continue to coexist with the overt pronominal form even though it does not surface in their productions during the intermediate and the advanced stages of L2 acquisition—that new parameter settings cannot be acquired in SLA. This sort of argument is consistent with the No Parameter Resetting Hypothesis (Hawkins and Chan, 1997; Smith and Tsimpli, 1995), which assumes L2 learners cannot construct grammars incorporating parameter values that are not realised in their L1 because learners only have access to UG via their L1 core grammars. This claim and approach may receive further support from the results of L1 French speakers learning L2 English if we agree with the common view that French is a non-null subject language. 128 That is to say, because their L1 and L2 share the same value
128French is commonly considered to be a non-null subject language, but it has subject clitics. Subject clitics in French have long been a subject of discussion and debate among linguists (e.g., Kayne, 1975; Auger, 1994; Borer, 1984; Jaeggli, 1982; Zribi-Hertz, 1994).
Basically there are two different approaches to account for their syntactic nature: the cliticization approach and the affix approach. Under the cliticization approach, which considers French as a non-null subject language, subject clitics are base-generated in
of the parameter [-null subject], the French learners managed to fix the parameter eventually in response to L2 input.
This view, however, contradicts the second scenario, namely the Parameter Resetting Hypothesis (Schwartz and Sprouse, 1996), which maintains that new parameter values can be reset as IL grammars are fully UG-constrained. Hence, restructuring is predicted to occur at the end-state grammars of L2 speakers, even in cases when the L1 and L2 differ in parameter values. According to this view, based on the observed gradual improvements in the learners’ performance at different proficiency levels sharing the same L1, one could argue that they would readjust their L1 value of the null subject parameter to the value appropriate to their L2 with increased proficiency in L2. In other words, because the advanced-level learners have a greater command of the target grammar under investigation compared to their less-proficient upper-intermediate-level peers, who themselves outperform their less proficient lower-intermediate-level counterparts of the same linguistic background, it could be argued that convergence to native-like performance will eventually come.
However, this argument seems to contrast with the results obtained in particular from the advanced-level Finnish and Arabic speakers, which indicate that their IL grammars still diverge from the grammar of English native speakers even though they are in the advanced stage of L2 development. However, this contrast can be explained by the fact that it is extremely difficult to determine if an L2 learner is really at the end-state of the developmental process; a learner might be somewhere in the advanced level of proficiency, yet not at the end-state (see Lardiere, 1998a, 1998b; Long, 2003; and White, 2003 for relevant discussion).129 To
the canonical subject position within VP, and then undergo movement to their preverbal position. However, under the affix view, which considers French as a null subject language, subject clitics are considered to be agreement prefixes. For reasons of space limitations, I will not comment on any particular position in this thesis. I will go with the common assumption that French is a non-null subject language. However, the interested reader is referred to the references above; or for a review, he/she is referred to Prévost (2009).
129 See also the methodological problem discussed in section 4.3.5 in relation to measuring L2 learners’ levels of language proficiency. Despite the satisfactory solution proposed to solve this problem, it should be added at this point that the only way to
illustrate this notion, the advanced-level learners’ data need to be analysed at the individual level to see if there is some individual variation in performance within the same L1 group. In fact, difference in performance is predicted on an individual basis in the light of the maximum and minimum number of null subject acceptances reported in the subgroups’ descriptive analyses presented in subsection 5.2.1.1 above. For example, Table 5.3 showed that the advanced-level Finnish participants’
acceptance ranged from zero to three sentences with referential null subjects on an individual basis.130 Figure 5.9 precisely illustrates the differences in performance among every individual advanced Finnish L2 learner of English.
Figure 5-9. Number of sentences with null subjects accepted by every individual Finnish advanced participant.
Figure 5.9 shows that 30 out of 53 Finnish advanced-level participants performed like English natives; they accepted no ungrammatical sentences. The
determine if a learner indeed reached steady-state grammar is by means of longitudinal data. This view is also stated in Lardiere (1998a, 1998b).
130 Due to the large number of participants and space limitations, only the descriptive results from Finnish advanced-level participants are presented at the individual level.
other 23 participants accepted sentences with null embedded subjects. While 14 out of those accepted only one sentence with mistakes, seven participants incorrectly accepted two sentences. The other two participants each accepted three sentences with mistakes. Therefore, it could be argued in agreement with the Parameter Resetting Hypothesis that because the majority of the participants converged to native-like usage, the others need more time to reset the parameters as they have not yet reached the end-state.
The discussion so far has shown that these results must be interpreted with caution. Even though the second scenario does not explain the discrepancy between acceptance of null subjects and the absence of null subjects in the learners’
production, especially at the advanced level of proficiency, the explanation based on differences in the task demands would account for such discrepancy/inconsistency only at the early stages of the acquisitional process. Therefore, I will further analyse the data before drawing a conclusion about the issue of null subject parameter resetting. Until then, this issue will be left open; I will return to it after discussing how the learners’ results diverge across the different syntactic formations under investigation. This will be the focus of the following section of the chapter.