Developing Processability hypotheses for the acquisition of Italian
Stage 1 Words (undifferentiated)
3. The role of word structure
1.4 Processing constraints on L1 transfer
Pienemann, M., Di Biase, B., Kawaguchi, S. and Håkansson, G. (2005). Processing constraints on L1 transfer. In J. F. Kroll and A. M. B. de Groot (eds.), Handbook of bilingualism: Psychological approaches, pp. 128-153. Oxford and New York: Oxford University Press.
This is a very long paper dealing with several languages which are typologically quite different. For reasons of space, only some parts of the published paper are reproduced below: (i) the abstract, in order to give a general overview of the original, (ii) my own specific contribution regarding the acquisition of Italian, and (iii) the conclusion, because I contributed to the general organisation of the paper and discussed with other co-authors their own specific sections. My own contribution here argues against the “full access/full transfer” hypothesis proposed by Schwartz and Sprouse (1994) and against parametric approaches such as White‟s (1989). Given the typological nature of the argumentation of the whole paper, Italian, as a pro-drop language, is significant in showing that learners latch on early to the canonical word order of their L2 even when it is different from their L1. Learners of Italian produce null subject sentences from the beginning of their learning, even though their L1 (English) requires obligatory subject. This is because „canonical order‟ coincides, universally, with the first L2-specific stage (category procedure), according to PT.
This article, as well as making a strong case for developmental constraints on transfer, clearly states, among other things, PT‟s position on „canonical order‟, anticipating the „Unmarked Alignment Hypothesis‟ explicitly formulated in the new PT extension (§ 2.2). For the learner, the initial hypothesis for word order coincides with the most prominent word order in the L2 – the default pattern in that language. This requires no exchange of grammatical information within the sentence because it relies on direct mapping of semantic roles onto surface structure. Consequently it has relatively low processing requirements. This explains why a learner of Japanese L2 can produce perfectly grammatical sentences very early in their learning career (see section on Typological Distance, at no disadvantage to the original paper, not reproduced here, pp. 145-146).
Italian, however, unlike Japanese, has a complex pattern of Subject- Verb agreement, so it is highly unlikely that learners of Italian will
produce grammatically correct sentences at Stage 2 except by chance or by chunk. Indeed, they don‟t, producing verbs most often with the „wrong‟ ending. On the other hand, if “full access/full transfer” was an option for learners, then those with an L1 characterised by obligatory subjects should not take the liberty of dropping them, especially at the beginning (i.e. the last stage of their L1) – but they do drop them and rely on contextual and pragmatic cues, as native speakers also do, to identify person deixis, even though they cannot rely on morphological cues. As a matter of fact, the learning of the full morphological paradigm of the six Italian persons will only happen later rather gradually (hardly a parametric event)
This theme, central to Italian, is further developed in the introductory section to this Chapter 1, while morphological development is the dominant theme of the previous three papers (cf. §§ 1.1-1.3). The next chapter, on the other hand (Chapter 2), will expand on the development of syntax in Italian L2, venturing beyond the boundaries of original 1998 PT.
Pienemann, M., Di Biase, B., Kawaguchi, S. and Håkansson, G. (2005). Processing constraints on L1 transfer. In J. F. Kroll and A. M. B. de Groot (eds.), Handbook of bilingualism: Psychological approaches, pp. 128-153. Oxford and New York: Oxford University Press.
Summing up, we find that all learners of Japanese studied longitudinally by Kawaguchi start with SOV word order and with subject omission, although their first languages follow an SVO pattern and one of the L1s does not permit subject omission. Obviously, these findings falsify the hypothesis that first language features are transferred to the second language at the initial state.
This raises the question as to why second language learners would start out with a structure that is typologically rather distant from their first language. The answer is implied in PT, and more specifically, in the ‗developmentally moderated transfer hypothesis‘ advocated in this paper. In relation to the initial hypothesis for word order, PT predicts the use of a canonical word order pattern. Japanese follows a canonical SOV word order, which requires no exchange of grammatical information within the sentence as it can rely on direct mapping of semantic roles onto surface structure (cf. our discussion above). In other words, because of the low demands on processability, this word order pattern can be processed at the initial stage of clause development despite the typological distance between the first and the second language; for a more detailed and formal account of information distribution in Japanese syntax see Kawaguchi (2005) and Di Biase and Kawaguchi (2002).
This analysis of the initial word order in the acquisition of Japanese as a second language also highlights a key difference between Clahsen‘s (1984) strategies and the processability approach. As Vainikka and Young-Scholten (1994) and Towell and Hawkins (1994) point out, Clahsen‘s strategies would predict that the initial hypothesis in L2 acquisition is formed on the perceptual array ‗actor, action, acted-upon.‘ thus producing universal SVO patterns for all L2s. No such assumption is made in PT. The only stipulation that exists at this level is that no grammatical information be exchanged within the sentence. This constrains the language processor to produce only structures that can be processed without such information exchange. SVO and SOV both satisfy this condition.
Di Biase (in preparation) has studied another typological constellation of the same kind as Kawaguchi. In his study he focused on the acquisition of a pro-drop language11 (Italian) as L2 by speakers of a non pro-drop language (English). According to White‘s (1989, 87) analysis, this type of learner has to learn two things: 1) the fact that null subjects are permitted, and 2) the circumstances in which the language makes use of null subjects. These assumptions are derived from the more general assumption that L2 learners transfer the setting of the L1 parameter to the L2.
Di Biase (in preparation) carried out a longitudinal study with two Australian informants over a one-year period. Both informants were university students of Italian. One informant (Ernie) was a beginning learner who had had no previous exposure to the language. The first set of data (= t1) was collected as soon as the learner started producing utterances with more than one constituent. The second informant (Lisa) was an intermediate student when the first set of data was collected.
11
A language in which the co-referential grammatical subject does not need to be pronominalised.
Table 9 compares the three different types of realisation of grammatical subjects, null, pronominal and referential, in both informants. It is easy to see that, contrary to White‘s prediction, both learners start with a high level of null realisations of grammatical subjects. In fact, the level of null-subject realisations found in these data is not unlike that found by Bates (1976) in her study of Italian first language acquisition, namely about 70%.
In terms of the processability hierarchy, null-subjects are placed at the same level as pronominal subjects because both are directly derived from c-structure (for a detailed analysis of information distribution in these structures see Di Biase and Kawaguchi, 2002). Therefore Di Biase‘s finding supports the prediction that the acquisition of a typologically distant L2 does not necessarily cause a learning barrier as long as the structure in question is located at the lower end of the processability hierarchy – even if it does not exist in the L1.
This finding is supported by other studies with the same typological constellation as that in Di Biase (in preparation). For instance, Phinney (1987) studied the acquisition of null subjects in English learners of Spanish (i.e. L1 = - pro-drop, L2 = + pro-drop) and also found an early appearance of null-subjects. Liceras and Diaz (1999) studied the acquisition of L2 Spanish (+ pro-drop) by speakers of Chinese, English, French, German and Japanese (i.e. speakers of both types of languages) and found a consistent early appearance of null-subjects in all informants.
Summary and conclusion
In this paper we proposed that L1 transfer is developmentally moderated. This hypothesis follows from the internal design of PT, which provides a framework for relating specific L1 and L2 structures to a universal hierarchy of processability, based on grammatical information transfer in the production process. This overall framework predicts which processing procedures are required for the processing of specific L2 structures. This is the basis for the general prediction that L1 knowledge and skills can be utilised for L2 processing only if the necessary processing resources have developed. Our developmentally-moderated-transfer hypothesis was tested empirically in the context of both typological proximity and typological distance. We demonstrated that, in both types of context, transfer may or may not occur. The key predictive factor is always processability. In other words, processability acts as a constraint on L2 transfer
and may override typological distance. In addition to this constraining effect, processability also has a facilitating effect which sets in (given structural L1 – L2 overlap) once the L2 has developed to the point at which the L1 structure is processable. This is evident in the advantage of Turkish learners of German over Russian learners of the same L2, with respect to word order (cf. Haberzettl, 2000).
The empirical studies discussed in this paper also shed light on the validity of the set of competing approaches to L1 transfer discussed above. We can now state that the ―full access/ full transfer‖ hypothesis by Schwartz and Sprouse (1994) is strongly falsified by all cases of non-transfer reported above, since the latter authors assume that the final state of the L1 is the initial state of the L2. The falsification of this assumption is particularly obvious in the study by Håkansson et al. (2002), which shows that Swedish learners of German do not transfer verb-second, although both languages contain this structure.
The so-called ―minimal tree hypothesis‖ predicts that L1 word order is transferred to L2. This is falsified by Kawaguchi‘s (in preparation) observation that Australian learners of Japanese start with an initial SOV hypothesis. This observation and the study of Swedes learning German both also falsify the transfer hypothesis implied in the Competition Model (cf. MacWhinney, this volume) according to which all transferable structures will be transferred at an early stage. The above studies show that this prediction is not borne out by empirical data. Swedish learners of German do not transfer V2 to German (which would yield a correct result) and Australian learners of Japanese do not transfer SVO to Japanese (which would yield an incorrect result). The strong initial transfer assumption inherent in MacWhinney‘s (1997, this volume) competition model also produces predictions which are falsified by empirical data, particularly by the Swedish-German study (Håkansson, Pienemann and Sayehli, 2002), which shows that verb-second is not transferred from Swedish to German even though this structure exists in the L1 AND in the L2. All other cases of non-transfer discussed above prove the same point.
In addition, it may be useful to consider the explanatory parsimony of MacWhinney‘s assumption that ―...all aspects of the first language that can possibly transfer to L2 will transfer‖ (MacWhinney, 1997, 119). The reader will recall that MacWhinney (1997) illustrates his point about structurally ―impossible transfer‖ using German and English as an example. German nouns are implicitly marked for grammatical gender whereas English nouns are not. He concludes that German learners therefore have no basis for
transferring the German gender system to English. Therefore this set of features is not included in his list of things that will be transferred.
Our point is the following. Whereas L1-L2 contrasts are transparent to the linguist, the question remains as to how the learner recognises these differences. The reader will recall that at the beginning of this paper we argued that the relationship between German and English diacritic features (of nouns) is not obvious to the learner and that a full transfer hypothesis would lead to unwieldy hypotheses. Conversely, it is precisely this lack of transparency in the relationship between L1 and L2 that makes a radical no- transfer hypothesis equally unlikely.
Assuming a lexically driven model of language production such as the one proposed by Levelt (1989), gender is one of several diacritic features residing in the lexical entry for (German) nouns, and the learner will have to discover for all lexical classes (such as noun, verb etc.) which of the L1 diacritic features are also marked in the second language, using known or unknown linguistic means, and which additional diacritic features are marked using which linguistic means. This is a monumental learning task. Assuming that diacritic features such as ‗gender‘ are not transferable for structural reasons would amount to a classical conditioning assumption within the competition model, which would assume a strictly linear relationship between input and output, following the motto ―if it is not in the input it cannot occur in the output.‖ As noted above, empirical data falsify such an assumption. This is also illustrated by the well- attested example of over-generalisation in English regular past marking, such as in Cazden‘s ―She holded the baby rabbits...‖ (1972, 96).
As these examples show, the assumption of a strictly linear relationship between input and output and a rich transfer assumption produce predictions which are falsified by empirical data – at least for the domain of morphosyntax. A rich transfer assumption is not supported in the area of bilingual first language acquisition either. According to De Houwer (this volume), no studies have empirically backed up the existence of the sort of language repertoires that would be predicted to develop in bilingual children in line with a transfer theory. Indeed she maintains that the interpretation of morphosyntactic features of the two input languages would assume that processing mechanisms in bilingual children would enable them to ―approach each input language as a morphosyntactically closed set.‖
The gist of the cross-linguistic survey of L1 transfer presented in this chapter can be summed up in two fundamental trends:
(1) Structures higher up the processability hierarchy are never transferred at the initial state – regardless of typological constellation.
(2) Initial word order may vary as long as the flow of grammatical information is restricted to the initial stage of processability.
These trends clearly contradict any theory that places emphasis on extensive L1 transfer at the initial state and support a view of transfer that is sensitive to the developmental state of the learner‘s language.