The articles that constitute the body of this journal issue present rich empirical evidence and theoretical reasoning that relate to the acquisition of relative clauses (RCs). Most are concerned with SLA, more specifically with the acquisition of East Asian languages (in particular, Japanese and Korean) by those who are already speakers of other languages, including European languages. The article by Yip and Matthews (this issue) is slightly different in that it documents the bilingual acquisition of Cantonese and English by children, with English being acquired with a slight lag relative to Cantonese, but, like the other articles, it also provides evidence of the confrontation of European and East Asian languages in the general area of language acquisition.
A uniting theoretical concern across the articles is the relevance of the noun phrase accessibility hierarchy (NPAH; Keenan & Comrie, 1977), although in discussing this concept—in part for independent reasons concerned with the observed patterns of acquisition—a number of other theoretically interesting notions also play a major role in the development of the authors' ideas. In my commentary, I will concentrate on three of these notions: first, and perhaps inevitably given my own role as its cocreator, the NPAH; second, the claim that at least some East Asian languages, most notably Japanese, have a general noun-modifying construction that subsumes translation equivalents of RCs (Matsumoto, 1997), an idea that I have espoused in my recent thinking on the typology of RCs (Comrie, 1996, 1998a, 1998b); third, the relevance of internally headed RCs, which turn out to play a much more important role than one might have predicted in these studies of SLA.
However, a couple of general observations are in order. By observation, I mean almost literally that; these are points that seemingly inevitably occurred to me as I read through the articles.
First, a question that almost invariably occurs in SLA is the extent to which learners carry over features from their native language into the language they are trying to acquire. What strikes me here is that, for native speakers of European languages learning Japanese or Korean as a second language (L2), the articles in this collection suggest surprisingly little transfer—an observation that is surely significant on a par with the dog that did not bark in the night. Indeed, speakers of European languages seem quite happy diving into such exotica as internally headed relative clauses (RCs) and forgetting whatever constraints on extraction their native languages might have led them to carry over to the L2 learning task. Because speakers of European languages are likely to carry over features of their own language in learning other European languages, as is apparent from my own observations, the low level of transfer in the acquisition of Japanese and Korean cries out for an explanation, and I hope future work will address this question in a more systematic way than I can do here. My current thoughts are probably little better than speculation, but one relevant factor might be that RCs in Japanese and Korean are so different structurally from their European translation equivalents—differing, for instance, both in constituent order (prenominal vs. postnominal) and, in general, in the absence versus presence of overt clues to the role of the notional head noun in the RC—that learners might simply treat them as a new structure to be learned rather than as variants of any structure familiar to them from their native language. (The Yip and Matthews contribution does provide evidence of transfer from Cantonese to English, and although the situation—bilingual acquisition, albeit with it a lag between Cantonese and English—is different, one again wonders why the pattern should be so different here. Does English also fail to provide enough overt clues as to what is going on in its syntactic structure?)
Second, one must nonetheless recognize that SLA involves a number of factors other than just the structure of the language being acquired. Even if the structure of the native language seems not to be particularly relevant in most of the data under consideration in this thematic issue, there are still other factors to be considered in stages preceding the (possibly never attained) nativelike knowledge of the construction in question, such as universal patterns of SLA or general problem-solving strategies that might be used by L2 learners. Indeed, the articles provide evidence for and explicit discussion of such instances. However, this means that one has to be careful in drawing any conclusions about the structure of language X from the way in which speakers of language Y acquire language X. In such cases, we might be looking directly into the structure of X, but there is also the possibility that we are looking at some version of X distorted through the prism of a learner of Y (whether this involves transfer or some other factor). In other words, we can use features of the language being acquired to explain properties of the SLA process, but it would be dangerous indeed to use properties of the SLA process to reconstruct features of the language being acquired. In any event, one hopes that there would be more direct ways of investigating the structure of that language!
Third, I am pleased to see a careful distinction made between production and comprehension of RCs where relevant in the articles that constitute this special issue. Especially given the fact that RCs in East Asian languages, in comparison with European languages, give the hearer fewer explicit clues as to what is going on (and, in particular, to what position is being relativized), one might expect that speakers of a language accustomed to such cues would have difficulties in comprehension that are not matched, or are not matched to the same extent, in production, and this prediction is borne out by some of the studies that make up the present issue. In particular, the greater difficulty provided for comprehension by reversible examples (e.g., where both interpretations “X V-ed Y” and “Y V-ed X” are plausible) suggests that comprehension is often driven more by pragmatics, with the learner making sense of the utterance with minimal or no appeal to syntax and therefore coming unstuck precisely when syntax is crucial to distinguishing different utterances.
NOUN PHRASE ACCESSIBILITY
To discuss the relevance of the NPAH to the L2 data, it is important to have a sufficiently clear idea of what the predictions of the NPAH are. I will give a brief, largely intuitive account that might differ in some details from Keenan and Comrie's original 1977 publication but that corresponds both to my own more recent thinking on the NPAH and, in general, to the way in which the notion has been received and interpreted in the broader linguistic community. The essential intuition is that there is a hierarchy of grammatical positions—those for which there is the strongest evidence being subject > direct object > other object > genitive/possessor—that defines a hierarchy of increasing difficulty of relativization, such that a position lower on the hierarchy is more difficult to relativize than one higher. (By “relativizing on” a position I mean, consistently, that the position in question is that occupied by the notional head of the RC within the RC; I do not refer to the position of the head in the matrix clause.) This finds two important realizations in linguistic data.
First, in language-specific terms: If a language can relativize a particular position on the NPAH, then it must also be able to relativize all higher positions. Thus, if a language can relativize direct objects, it must also be able to relativize subjects. Note that a language that can relativize all positions on the NPAH is consistent with this generalization, although clearly if all the world's languages could relativize all positions on the NPAH, then the NPAH would be a singularly unrevealing contribution to expressing this. Crucial to the validity of the NPAH is the second, crosslinguistic (or typological) characteristic of the NPAH: Each position on the NPAH is justified empirically by the existence of at least one language that relativizes on that position but on no lower position. This typological statement is crucial to the validity of the NPAH because it is this that provides the empirical evidence for each of the positions on the NPAH. However, it is nonetheless important to note that a language that happens to allow relativization on all positions on the NPAH is still consistent with the NPAH. In other words, if a language fails to provide evidence for the NPAH and, equally, if the interlanguage used by some speaker of language X acquiring language Y fails to provide such evidence, this is not evidence against the NPAH or even that the language in question is driven by some force other than the NPAH—although this might be the case, it would have to be established in positive terms, not just in terms of the absence of NPAH effects.
It is equally important to clarify exactly what is meant by a NPAH effect and, perhaps even more importantly, how one assesses the relevance of such effects. If there is a clear difference between accessibility to relativization of, say, subjects and direct objects in a language, then, in one sense, this is a clear instance of a NPAH effect. However, what governs NPAH effects? In Keenan and Comrie (1977), the authors were deliberately noncommittal about this, and even almost 30 years later, I would still be reluctant to commit myself, in part because this would go well beyond my own expertise. Although the NPAH has often been interpreted as claiming that grammatical relations determine accessibility to relativization and although some unfortunate formulations in the 1977 article can be interpreted as claiming this, it is clear from elsewhere in the article both that other factors can intervene—that is, any linguistic principle must be seen in its interaction with other linguistic principles—and that there is no claim that the grammatical positions (which are for the most part also grammatical relations) are the primitives that drive accessibility. Indeed, there are hints in the 1977 article that the NPAH might be a reflection of more fundamental psycholinguistic principles—an idea that has been developed in other work (see, for instance, the commentary by J. Hawkins, this issue, and the references cited therein).
This idea becomes particularly important when one moves from the relation between the NPAH and grammaticality judgments to relations between the NPAH and various aspects of linguistic performance, including SLA. Indeed, in Keenan and Comrie (1977), there is already allusion to the possible relevance of the NPAH to first language acquisition, with reference to studies that show differential ease of acquisition of positions lower and higher on the NPAH, but it is important to note that the phenomenon in question here is differential ease of acquisition, not the clear-cut differentiation that one would at least hope to find in grammaticality judgments. Also, when it comes to differential occurrence of relativization on positions higher and lower on the NPAH in texts judged independently to be stylistically more and less difficult, one is clearly not dealing with the possibility versus impossibility of relativizing on a particular position but, rather, with evidence from the frequency of relativizing on different positions. All of this suggests that whatever the ultimate nature of the factors underlying the NPAH, they go well beyond their realization in grammaticality judgments. In particular, performance in a particular language might show evidence for the NPAH that is not directly reflected in grammaticality judgments on relativization on particular positions on the NPAH in that language.
As will be discussed in more detail in the next section, the claim has been made that some East Asian languages (in the first instance, Japanese) do not have a distinct RC construction but, rather, a general noun-modifying clause construction, such that this construction is interpreted by trying to find a semantically and pragmatically plausible relation between the modifying clause and head noun (Matsumoto, 1997). It follows from this that such languages should lack syntactic constraints on accessibility, although they might (and, at least in the case of Japanese, clearly do) have quite restrictive semantic and pragmatic constraints. What predictions would one make for such a language in terms of the NPAH? If the NPAH were to be interpreted as a pure syntactic determinant of grammaticality judgments, then the conclusion would inescapably be that the NPAH should be irrelevant for such languages. However, if one regards the NPAH as being a reflection of more basic underlying principles, some of them psycholinguistic in nature, then there is no reason why those more basic underlying principles should not manifest themselves in a language with a general noun-modifying clause construction. In other words, it is not the case that the NPAH predicts that the acquisition of a L2 with a general noun-modifying construction should not show NPAH effects, where NPAH effect is defined simply as differential ease of acquisition of different positions on the NPAH, although ultimately such differences would have to be accounted for in terms of whatever more abstract principles underlie the NPAH. In this regard, it should be noted that Keenan and Comrie (1977) said nothing about how the ease of relativizing different positions on the NPAH should correlate with other general syntactic properties of RCs, such as whether it is conditioned primarily by syntactic extraction.
Just as one cannot directly go from the NPAH to predicting what will happen in the acquisition of a L2 with noun-modifying clauses, equally the NPAH does not allow one to deduce from properties of the SLA of a particular language whether this language has a general noun-modifying clause construction. In particular, complete absence of NPAH effects is, as noted earlier, fully consistent with the NPAH and could simply reflect the trivial instantiation of a member of the set of languages allowed by the NPAH in which relativization is permitted on all positions in the NPAH.
Notwithstanding this situation, it is clearly a significant result of the articles in the present collection that there is a low level of evidence for NPAH effects overall. This might correlate with the fact (assuming, of course, that it is a fact—see the following section) that the languages in question have the general noun-modifying construction—although note that internally headed RCs would also be expected to give a similar result because there is no reason to expect syntactic constraints on which positions can be relativized using this construction—and, indeed, provides good circumstantial evidence in favor of this assumption. However, crucially, it is only circumstantial evidence and, ultimately, other possibilities would need to be excluded. The material and discussion provided in the articles in this issue provide an excellent basis for the further development of this program.
It is a basic strength of the contributions to this issue that they strive to find other, non-NPAH explanations for the major patterns observed. These explanations typically revolve around features of the language that provide speakers of particular first languages with a nice entry point into the L2 construction. For instance, relativizing on a direct object in a language like Chinese, given its constituent order properties (SVO clause order, prenominal RCs), means that a RC with the order [SV]-head will have the same order as a clause with the structure SVO, whereas relativizing on a subject will give a structure like [VO]-head, which cannot be readily mapped onto a simple clause structure. This is an independent factor that can lead to greater ease of acquisition of object relativization than of subject relativization, irrespective of whatever predictions one might want to derive from the NPAH. Language, in both structure and performance, is an interaction of various principles, and when these principles pull in different directions, one principle might well have to yield.
NOUN-MODIFYING CLAUSES
As noted in the previous section, the claim that (some) East Asian languages have a general noun-modifying construction should not be translated literally into a prediction that those languages will show no NPAH effects, let alone into the claim that L2 learners of those languages will show no NPAH effects (given that the properties shown by L2 learners are not simply properties derived from the structure of the L2, as noted in the first section of this commentary), although there might well be a plausible presumption that languages with and without general noun-modifying clause constructions will behave differentially with respect to NPAH effects. However, clearly, the relevance of this observation will depend on whether the language in question really does have a general noun-modifying construction. This analysis of the relevant Japanese construction seems to have become by and large (although not universally) accepted among proponents of both functional and formal approaches to syntax. As the contributions to this issue, in particular that of Jeon and Kim, show, extension of this analysis to other East Asian languages is by no means uncontroversial, so it is perhaps worth spending some time to clarify the issue.
As noted in Comrie (1998b), the question of whether a particular language does or does not have general noun-modifying clause constructions is one that must be argued in detail for each language because a superficial examination might fail to reveal the crucial distinctions that are involved. In Korean, I take the crucial evidence to be that relativizing the genitive/possessor position requires a resumptive pronoun. (Other claims, such as that relating to empty pronouns, are, of course, not evidence but, rather, demonstrations that an extraction analysis is not incompatible with the relevant data. In other words, the relevant data are neutral with respect to the opposition between extraction and noun-modifying clause analyses.) I have no vested interest in claiming that Korean has a general noun-modifying clause construction, and although I have claimed this in the past (Comrie, 1996, 1998a), I am quite happy for the field to move forward by proving me wrong. However, I think it is important to understand the argumentation that would be needed to distinguish clearly between the two positions.
First, it should be noted that the noun-modifying clause analysis is not a claim that anything goes. Because the position relativized must be retrievable semantically (from the frame of the predicate; Matsumoto, 1997), pragmatically, or both, this can actually place heavy constraints on what can be relativized and—as noted by Matsumoto—many RCs that are perfectly natural in English are excluded in Japanese, precisely because there is no plausible way of combining semantics and pragmatics to arrive at a plausible interpretation. In some instances, the result might even be the categorical unacceptability of relativizing a particular position. For example, because objects of comparison never form part of the frame of a predicate, one would expect, other things being equal, that relativizing objects of comparison would be impossible in a noun-modifying construction. (The hedge “other things being equal” is designed to allow for languages in which there might be some other overt marker indicating that an object of comparison is present. This is not the case, for instance, in Japanese.)
Jeon and Kim (this issue) noted that in Korean, the requirement of pronoun retention for relativization on genitives is categorical, contrasting with the general principles that govern the occurrence of pronouns in Korean, which are rather discourse-driven. The point is well taken and puts the onus on those who would continue to defend a noun-modifying clause analysis. However, at least it is clear in what direction such a defense would have to go: It would need to show that the discourse principles that govern the occurrence of pronouns in Korean predict that the occurrence of a pronoun should become categorical in the case of what is interpreted as relativization of a genitive. This would presumably require detailed analysis not so much of pronouns in general but of genitive pronouns in particular. The challenge has been stated, and although I have no idea whether the necessary investigation will be easy or of the direction in which it will lead, at least the issue is clear.
It should be noted that recent work has independently pointed to problems with a clear-cut distinction between noun-modifying clause construction and traditional RCs. Colloquial English, for instance, quite readily allows noun-modifying clauses with the conjunction where, as in Dickens is one of the few authors where I'd rather watch the video. However, this construction with where is not possible where one would otherwise be relativizing a highly accessible position, such as a subject, as in *Dickens is one of the few authors where they achieved success in their lifetime. Clearly, there is a mass of investigation that still needs to be carried out to clarify the distinction and, in particular, to explain why a construction whose basic function is to link a clause to a head noun without specifying the relation is excluded with certain relations, as in the second English example.
INTERNALLY HEADED RCs
As noted in the first section of this commentary, perhaps my greatest surprise in reading the contributions to the present issue is the frequency with which the internally headed RC construction occurs—more specifically, in the interlanguage of speakers of languages that completely lack this construction, such as most European languages. Although we no longer view the internally headed RC as quite the exotic specimen it seemed to be three decades ago, with the construction having become widely attested in many parts of the world, it is nonetheless still a construction that, in its canonical form, seems alien to speakers of European languages. Indeed, whenever I have to construct an example of an internally headed RC, I usually still have to resort to a painstaking process that shares more with general problem solving than with usual language-related abilities. I am thinking of examples of the type where, as a translation of I will sing in the house that you built, one finds literally “I [you house built]-in will-sing,” where not only is the head internal to the RC, but, equally, the RC is internal to its matrix clause. The ease with which speakers of European languages fall into internally headed RCs in East Asian languages is even more surprising when one notes that this is far from the most frequent kind of RC in these languages and there are quite heavy restrictions on them.
Before succumbing to raw amazement one should note that under certain structural circumstances, internally headed RCs lose much of their exoticness. This occurs in particular if the internally headed RC occurs not in the middle of the matrix clause but at one end, usually the beginning—for instance, if the example given in the preceding paragraph were rather “[You built the house] I'll sing there.” Effectively, what one needs to do in order to process such an example is to process each clause independently and then relate them plausibly to one another, using the explicit or implicit shared noun phrase. The construction thus becomes similar to the so-called correlative RC, as found in Indo-Aryan languages with a structure like “which house you built, I will sing in it.”
It is interesting to note that the examples of internally headed RCs presented in the articles in this special issue are overwhelmingly, if not exclusively, of this kind and thus avoid the potential processing problems posed by an internally headed RC that is itself internal to the matrix clause. Indeed, for examples with the RC peripheral to the matrix clause, the processing problem is presumably less than for almost any other kind of RC construction, where at least some configurations require interrupting the processing of the matrix clause in order to deal with the subordinate material. Thus, what at first sign seems to be a surprising thrust toward complex exotic material on the part of speakers of European languages might turn out in fact to be a grasp at something that at last makes sense amid alien patterns of clause combining. If so, this is a nice example of how SLA can go in directions not immediately predictable either from the structure of the native language, or from the structure of the language to be acquired, or from their combination.
CONCLUSION
The wealth of empirical material, generalizations, and analysis presented in the articles that constitute the body of this issue is a major contribution to our understanding of the SLA of RCs and their translation equivalents. Indeed, the material is so rich that I am convinced it will provide fuel for many years of future discussion and investigations—the surest hallmark of a major contribution to the field.
