Appearance of Head-Internal Relative Clauses
Let us now turn to the other problem left open in Watanabe 2002. This problem is related to the fact, discussed in Watanabe 2004c, that head internal relative clauses (HIRCs) started to appear at about the same time as wh-movement was lost in Old Japanese.
There are at least two types of wh-in-situ languages. The type represented by Chinese and Lakhota mainly makes use of unselective binding. The Japanese type, on the other hand, forms the dependency by means of Agree. After overt wh-movement was lost in Old Japanese, the question is which type of wh-in-situ language emerged. Watanabe 2002 ends with this question, which is impossible to answer conclusively in diachronic syntax as long as we limit our attention to questions of locality. However, the difference between the two types is not limited to the mode of dependency. In Watanabe 2004c, I claim that the properties of wh-in-situ and the HIRC are closely connected. The type of wh-in-situ in Old Japanese can be identified by focusing on the HIRC and related properties.
Among the Chinese-type wh-in-situ languages, Chinese and Lakhota diverge in the availability of HIRCs. Given the structure in (1) for the HIRC (cf. Kayne 1994), its absence in Chinese and presence in Lakhota can be attributed to the absence (Chinese) or presence (Lakhota) of D.
(1)
The Lakhota HIRC is illustrated in (2a–b).
(2)
Note the definite article ki at the end of the (outer) HIRC in (2a–b). (2c) is an example of an ordinary clause, where the nouns are accompanied by an article.
It follows from the presence of HIRCs in Modern Japanese (see (3a)) that this language also has D. The properties of the HIRC in Modern Japanese di¤er from those of the Lakhota HIRC, however. While Lakhota can embed one HIRC inside another as in (2b) (Williamson 1987), Japanese cannot. Compare (2b) with (3b).
(3)
The head-D relation for the outer HIRC in (2b) and (3b), shown schematically in (4), is blocked in (3b).
(4)
This is an instance of a locality violation found with the Agree relation, but not with unselective binding. We are thus led to the conclusion that the dependency in HIRCs employs agreement in Modern Japanese and unselective binding in Lakhota.
This difference in HIRCs can be linked further to the type of indeterminate system found in each language (see Watanabe 2004c). As discussed in detail by Aoun and Li (1993) and Tsai (1994, 1999), the indeterminate system also has two varieties. Modern Japanese adds a particle to a wh phrase to form a non-wh quantificational expression, as in (5).
(5)
Chinese, on the other hand, does not add such a particle, as illustrated in (6).
(6)
Aoun and Li (1993) and Tsai (1994, 1999) indeed claim that this parametric variation of the indeterminate system is responsible for the difference in locality of wh-in-situ. Lakhota belongs to the Chinese type, as the following examples from Williamson 1984 show:
(7)
Recall that Lakhota patterns with Chinese in terms of the locality of wh in-situ.
How is the parametric variation of the indeterminate system related to the typology of HIRCs? The first step in answering this question is the idea, proposed in Watanabe 1992, that the quantificational particle attached to a wh-phrase is an instance of D as shown in (8) and that the wh-phrase in its wh-use is accompanied by a null D.
(8)
The common part is called an indeterminate. The absence of D in Chinese is consistent with the fact that the Chinese indeterminate system does not use particles. Now suppose that the quantificational particle attached to an indeterminate always requires agreement. The indeterminate-particle relation indeed exhibits the locality property typical of agreement, as observed by Takahashi (2002), who points to the following contrast:
(9)
Note that the configuration in (4) is found in (9), except that D agrees with an indeterminate in (9). (9b) is ruled out because the two D heads are of the same type. Suppose then that D in Modern Japanese is uniformly of the agreeing type. It follows that the head-D dependency in the Japanese HIRC involves agreement. The indeterminate system of the Chinese type, on the other hand, does not affect the nature of D since it is incompatible with D. Lakhota employs binding D outside the indeterminate system as a default. Chinese does not use D at all.
Parameterizing D in this way explains the properties of wh-in-situ as well. Notice that D is involved both in wh-in-situ (as the goal) and in HIRCs (as the probe) in languages with the Japanese-type indeterminate system, as in (10).
(10)
Since D is uniformly of the agreeing type, both wh-in-situ and HIRCs are sensitive to minimality. In languages with the Chinese-type indeterminate system, wh-in-situ does not make use of D and falls back on unselective binding by C, again as a default.
Table 1 summarizes the various properties of the three languages dis cussed here. D plays a central role in this typology. The key assumption is the uniformity of the D type in a given language.
Now we are ready to go back to Old Japanese. As observed by Kondo (1981), the HIRC started to appear in the tenth century, by which time overt wh-movement had already been lost. One of the earliest examples of the HIRC is given in (11).
(11)
Table 1
For more examples, see Kuroda 1974. As shown in Watanabe 2004c, the correlation between the loss of overt wh-movement and the appearance of HIRCs is strong evidence that the mechanism of wh-in-situ makes HIRCs possible.
As in the case of wh-in-situ, it is practically impossible to determine the type of the Old Japanese HIRC directly in terms of locality. If we look at indeterminates, however, it seems justified to conclude that wh-in-situ and the HIRC in Old Japanese are basically of the same type as their Modern Japanese counterparts, since Old Japanese also has an indeterminate system of the Modern Japanese type, as shown by the examples in (12).
(12)
In this period, mo is the only particle attached to an indeterminate. Redu plication as in (12a) was also productive at that time, judging from the quantitative data in Yamanishi 1987.
The conclusion that wh-in-situ in Old Japanese is the same as wh-in-situ in Modern Japanese now forces us to reconsider the analysis of the particle ka presented in Watanabe 2002, according to which it was the realization of an uninterpretable focus feature when the wh-phrase was overtly displaced. It turns out that wh-in-situ participates in feature checking after the loss of overt phrasal wh-movement in Old Japanese after all. There would be no reason why ka ceased to be used obligatorily after the loss of wh-movement, if ka had been the realization of an uninterpretable focus feature, the driving force behind agreement. What role, then, did the particle ka play in the seventh and eighth centuries?
