The study of relative clauses (RCs) has been central to both theoretical linguistics and studies in acquisition. This thematic issue of Studies in Second Language Acquisition makes a very important new contribution in this area by highlighting recent developments in thinking on the structure of RCs crosslinguistically and by introducing new data on the acquisition of Japanese, Korean, and Cantonese as first languages (L1s) and second languages (L2s). In this commentary, I will focus on two themes that emerge from the articles: representation and processing of RCs.
Current debates concerning representation in SLA focus on whether adult L2 learners are able to develop L2 representations that are nativelike (White, 2003). RCs are important in that they have complex properties that allow researchers to probe abstract knowledge of a computational system. The noun phrase accessibility hierarchy (NPAH) is a typological universal in that it holds true across a range of languages. However, one cannot emphasize enough, as Yip and Matthews (this issue) remind us, that such a typological universal can only be shorthand for a mental representation or processing challenge in the mind of each individual learner. Typological universals are useful and important when comparing L1s and L2s, but there is no other reason to consider them available to an individual except, perhaps, to determine whether a stage of development in an interlanguage grammar is a legitimate possible grammar in one of the world's languages. In fact, if processing is the reason that the hierarchy exists in one language, then the processing difficulty in that language must be addressed in the context of the all of the other properties of the grammar of a particular language. Properties that come to mind are pro-drop (both subject and object), grammatical particles and morphology, head direction, and the existence of overt and covert long-distance dependencies.
The authors of articles in this issue introduced significant suggestions by Comrie (1998) that question whether RCs have the same properties in all languages. It had been assumed that RCs in all languages have a similar structure, as illustrated with examples of English subject and object RCs in (1) and (2) where NP = noun phrase, IP = inflectional phrase, and CP = complementizer phrase.
The traces (ti) indicate a gap that is associated with the preceding filler. Such traces are assumed to be psychologically real and active. An additional assumption, following standard analysis, is that the gap is also coindexed with a position in the CP, which might be null (Gibson, Hickok, & Schütze, 1994). This detail is skipped over in most of the articles in this issue; however, intermediate traces are important in explaining why extraction of a wh-element out of a RC violates island constraints, as illustrated in (3) and (4). In (5), we see that long-distance extraction out of an embedded clause is grammatical.
According to the articles in this issue, Comrie proposed that head-final RCs in languages such as Chinese have the structure in (7) rather than that illustrated in (6) (see Yip & Matthews, this issue). If this is correct, then learners of such languages need only to acquire the position of the RC relative to the head, not the empty categories/operator structure that exists in a language like English.
The claim is that clauses such as (7) are attributive clauses that are loosely associated with the head NP semantically and pragmatically but not necessarily syntactically. This claim raises a question about the status of other clauses that seem to serve a similar function. In Chinese, two constructions appear to have the same function as RCs in English (Dai, 1987; Li & Thompson, 1981). The RC that precedes the head noun is the first type, illustrated in (6). In the second type, illustrated in (8), the RC appears to come after the head noun.
However, both Dai and Li and Thompson argued, on the basis of a variety of semantic and syntactic tests, that only the sentence in (6) contains a RC. The meanings of (6) pick out a person or a teacher as one of a set of individuals (i.e., they restrict interpretation), whereas (8) means “I have a sister and she happens to like American movies.” Dai noted that only RCs as illustrated in (6) provide a property for the establishment of a subset of people specified by the following NP (i.e., a restrictive RC), whereas (8) only provides an incidental property about the person denoted by the preceding NP. Finally, in (8), the NP meimei “sister” and the following clause do not form one constituent (Huang, 1987). The point of this discussion is that prenominal modifiers in Chinese must have some relation to the head noun that is different syntactically and not just semantically from structures such as (8). Andrews (2004) noted that several languages (e.g., Navajo) have a variety of modification structures that need to be considered when developing a typology of RCs. The articles in this issue do not specify what the syntactic relationships among such structures might be. Hence, this will also be a very fruitful area for future research.
Another comment about the grammatical status of RCs relates to culture and complexity. It is somewhat puzzling to a formal linguist that appeals are made to culture in the explanation of a syntactic phenomenon (Shirai & Ozeki, this issue). It is clear that culture plays a role in language in the structure and content of the lexicon. However, RCs have abstract properties that would seem inaccessible to the influence of culture and writing. Although the lack of context in written texts might encourage more frequent modification of NPs to assist identification of a referent, one has to remain skeptical that grammatical complexity is directly related to cultural sophistication. The articles in this issue do not discuss the tests and properties of RCs in very great detail—in particular, whether such clauses are islands and whether traces can be identified and how they might be shown to be independent of other NP modifier phrases such as adjective phrases. Once these relationships are established, the role of culture in determining complexity will be clearer.
With processing and representation comes the influence of word order and frequency. In this context, a comment on Chinese word order is appropriate. Nearly all of the articles suggest that Chinese is a SVO language (Mulder & Sybesma, 1992). It is true that unmarked objects follow the verb, as do the complements of matrix verbs. However, the ba construction, certain prepositional or postpositional phrases, and the subject/object drop properties of the language mean that the input in Chinese is unlikely to consist of mainly (S)VO structures. Certainly, dealing with corpora in Chinese turns out to be problematic because linguists do not agree on the analysis of some structures in Chinese (Xue, Xia, Chiou, & Palmer, 2005). (Incidentally, the same comment applies to Spanish, in which subject drop and preverbal object clitics create patterns that are OV rather than VO; VanPatten, 1996.) If animacy and frequency are really so important in the development of RCs, then more detailed corpus analyses, such as those mentioned by Ozeki and Shirai (this issue) and discussed in Kamimoto, Shimura, and Kellerman (1992) and Fox (1987), will be necessary to establish the exact nature of RC use.
The second question is whether adult L2 learners process the L2 using the same kinds of algorithms as native speakers (Clahsen & Felser, 2006; Fender, 2001; Juffs & Harrington, 1995). Such processing can be measured through comprehension, as in the study by Kanno (this issue), but such behavioral measures also need to be complemented through online processing studies. Although the articles in this issue test processing through comprehension and sentence combining (which are offline processing measures), none of the articles use online processing techniques such as self-paced reading (Juffs & Harrington), self-paced listening (Marinis, 2003, 2006), or eye-tracking. If the NPAH is truly a processing problem and not one of differences in representation, then evidence must be found both in offline and online processing measures in a variety of syntactic contexts.
In this regard, it is interesting to note that the subject-object asymmetry in English with RCs that modify a subject NP have rarely been found with RCs that modify object NPs, as in (9) (see Hamilton, 1995).
The fact that relatively little attention has been paid to such sentences in L2 research is surprising given the extensive L2 literature on this topic. It is interesting to note that Gibson, Desmet, Grodner, Watson, and Ko (2005) reported that RCs modifying an object were read more slowly than RCs modifying a subject. This result is surprising because, as they pointed out, the result contradicts the assumption that nested sentences are more difficult than those that are not center-embedded. This finding is rare and should be replicated. In this regard, most of the articles in this issue contain experiments with RCs without any other syntactic or discourse context. The strength of this approach is that it isolates RC types and reduces the influence of other factors. The weakness is that in natural speech, restrictive RCs identify a subset of potential referents in discourse, as noted in the natural production data collected by Yip and Matthews (this issue). By excluding such contexts, interpretation of the RCs might result in learners' knowledge being overestimated or underestimated. Future research will need to assess knowledge and processing of RCs in context.
To sum up, this issue is a very important first step in developing a comprehensive knowledge of how L2 learners represent and process RCs in East Asian languages. The articles point a clear way forward for future research that will establish the true nature of the internal structure of RCs in these languages and provide a base of offline behavioral data on which to develop studies of online processing and illustrate how processing difficulty of RCs might interact with other features of the grammar and discourse.
