Abstract
It is often the case in Japanese, or other languages, that some element of a sentence, such as an NP or a PP, is syntactically missing or barely specified but is semantically identified with an element inside or outside of the sentence. In Japanese, these phenomena are observed in a variety of grammatical constructions. For example, in reflexivization, an NP is only specified as zibun ‘self’ (without even indicating the gender, unlike English), but it is usually identified with the subject of the sentence. In causatives and intransitive passives, we find embedded VPs whose subjects are not explicitly specified (cf. chapter 3) but are semantically identified with the syntactic object.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Preview
Unable to display preview. Download preview PDF.
Notes
Hasegawa (198la) uses the term “control” in a similar sense to ours in her nontransformational treatment of control in English and Japanese, since she also assumes VP embedding and disallows a dummy.
In the following trees, the feature SUBCAT is abbreviated as SC to save space. In more precise presentations, which will be shown later, the feature REFL should appear in the nodes dominating the reflexive.
Note that both reflexivization and missing subjects are only concerned with control of a VP. The case where the controllee is a TVP is concerned only with missing objects and will be discussed later.
Here, the term direct argument is used in the sense where, in the function application form a(ß)(y) (= a(y, ß)), ß is the direct argument of a. Note that y is the direct argument of a (ß) but not of a.
Examples of missing subjects and objects will be presented shortly. Cf. (4.11) and (4.12).
This principle will be given a more formal presentation in terms of the current framework in section 4.4.
As in chapter 3, the parentheses in the English translation indicate that there is no overt counterpart in the original Japanese sentence.
The second argument of CAUSE is obtained by assuming first-order reducibility of the translation of sor ‘shaveü and 2PP(k) as the translation of Kenü:
As mentioned in note 37 in chapter 3, the SEM values will be generalized in chapter 6 so that proper argument binding is based on the SUBCAT value rather than lambda binding. In such a generalization, the SEM value for sase will incorporate object control more straightforwardly: Note that the object of sase and the semantic subject of the complement V (i.e., the PP[SBJ] in the SUBCAT value of the complement V) have the identical SEM value. Note also that the SEM value of the complment V, i.e., y, is itself embedded in CAUSE, since y is of type s in this generalized scheme. In the following, the reader has to keep in mind that the PP[OBJ] in the SUBCAT value of sase and the PP[SBJ] in the SUBCAT value of the complement V have the identical SEM value.
As Brame (1975, 1976) points out concerning the controversy over VP complements in English, and as is shown in the analysis of Japanese causatives and intransitive passives in chapter 3, assuming VP complements will eliminate the need for many of the familiar cyclic transformations such as tough-movement, Raising, Equi NP deletion, etc. This may not have been a welcome move for transformationalists since it reduces most of the motivations for having a transformational component in the grammar Note also that to construct compositional semantics for structures with embedded VPs is no problem in Montague semantics or the like by virtue of lambda abstraction. I am grateful to Gerald Gazdar for bringing Brameüs works cited above to my attention. See also Gazdar (1982b), which is a review of Brameüs more recent works concerning the controversy over VP complements.
There may be pragmatic or stylistic differences between gaps and explicit pronouns. In Japanese (as compared with English), pronouns are not often used for the purpose of mentioning the recurrent topic. Thus, if explicit pronouns are used, it may imply that the speaker is putting stress on the referent of the pronouns. These pragmatic implications are ignored in this book.
The only difference between the SLASH feature in GPSG and ours is that our SLASH feature takes as the value a set of categories, instead of a single category, as in GPSG. Since the analysis presented in Gazdar, Klein, Pullum, and Sag (1985) is concerned with English, this restriction on the value of SLASH may be more appropriate. However, as the examples in this chapter show, it is quite common to have multiple gaps in a sentence in Japanese (cf. note 42). The SLASH feature of HPSG (Pollard 1984, 1985) also takes a sequence of categories, and an analysis of English sentences that seem to have multiple gaps is presented using such a feature.
In the trees in this chapter, the feature SLASH is abbreviated as SL to save space.
This is a simplified version of the FFP in GPSG. See Gazdar, Klein, Pullum, and Sag (1985) for a more precise definition based on formal concepts.
Jacobson (1984) makes a distinction between “syntactically free gaps” and “linked gaps.” Her “linked gaps” seem to correspond to “bound” gaps here, since both typically appear in topicalization. Her “free gaps” are not gaps that originate from free instantiation of the feature SLASH; they cannot be lexically filled. Instead, they are also introduced by phrase structure rules of the form A - B C/D (where B s D). She uses “free gaps” to explain tough sentences in English.
The feature ADJUNCT, which is abbreviated as AD in the trees, specifies the category of the head that an adjunct adjoins; cf. note 15 in chapter 2. The FFP in the form of (4.25) applies only to complementation. As is shown in (4.28), in topicalization, the FOOT feature of the head S/PP, i.e., (SLASH {PP}), is not passed up to the mother, and hence the mother category is simply an S. In fact, sentences such as (4.26) are complete sentences on their own, with no missing subject or object. Since bound gaps and the FFP for adjunction will be discussed in more detail in chapter 5, we will concentrate on free gaps and complementation in this chapter.
We will define the term unify,which formally specifies matching of categories, in section 4.4.1.
In my judgment, the sentences in (4.29) sound more natural as Japanese sentences than those in (4.31). This seems to be a matter of style, not of grammar. Perhaps there is a stylistic preference (due to a Gricean principle) among speakers of Japanese to avoid superfluous use of proforms for obvious references; cf. note 11.
One of the reviewers points out that English allows a universal quantifier to be an intersentential antecedent of a pronoun. The example sentence given is: (i) Each student must obtain a registration form. He should fill it out and turn it in right away. This kind of use of (implicit or explicit) pronouns in Japanese seems to be rare, if it exists at all. Note the similarity of the it above and that in the celebrated ‘donkey sentenceü (Geach 1962): (ii) Every man who owns a donkey beats it. As is known, such ‘lazy pronounsü are problematic to the analysis in terms of logical formulas and hence seem to be out of the scope of the current analysis. For a promising way to handle these, see Kamp (1981).
The pronoun darenimo ‘no oneü is the dative form of daremo, into which the dative marker ni is morphologically incorporated.
If the object is not one of these, there is still a possibility for pragmatic control. Thus, (i) below could mean that Naomi said, “You forgot to put sugar in the cake.” (i) Naomi-ga Ken-ni keeki-ni satoo-wo irewasureta-to itta. NOM DAT cake-in sugar-ACC forgot-to-put-COMP said Naomi told Ken that she/he had forgotten to put sugar in the cake.
There may be more to be said about ‘zeroü pronouns. Since the problem of zero pronouns is one of the most actively studied topics in Japanese linguistics, there have been numerous recent works on putative syntactic and pragmatic constraints on these pronouns (e.g. Kuno (1978b, 1986a), Hasegawa (1984, 1985), and Kameyama (1985) among others), to which we have nothing to add. Thus, the generalizations in (4.36) represent only the most minimal explanation.
In GPSG (Gazdar, Klein, Pullum, and Sag 1985), the feature NULL is used to specify that the category immediately dominates a null string. It is always associated with a SLASH feature by the following Feature Co-occurrence Restriction: (i) (NULL +) (SLASH a), where a is not empty The NULL feature is introduced by a metarule (SLASH Termination Metarule 1), and hence some restriction applies in generating a gap in English I assume that gaps in Japanese are freely generated as lexical items, reflecting a difference between English and Japanese: the latter allows free gaps, while, in the former, gaps only appear in unbounded dependency constructions and are always bound.
Since control of the reflexive and gaps is optional, the CP gives a possible control structure only where the two PPs unify. When syntactic control does occur, the two PPs unify, and the PP in the SUBCAT value of M syntactically controls the other PP. Since obligatory object control is expressed as part of the semantics of particular lexical items, (4.39) only covers optional control.
As will be seen later, apparent object control of the reflexive can also be explained by the CP without any modification.
In complementation — M - C H — the SUBCAT Feature principle, (2.13), specifies that the SUBCAT value of the mother is that of the head minus C, which unifies with one of the categories in the SUBCAT value of the head. Thus, what is not passed up to the mother from the head unifies with C.
The complementizer to is assumed to be a P that subcategorizes for a sentence in the current analysis. More specifically, its lexical entry is the following:
The decision to treat syntactic control in terms of the SUBCAT feature rather than SLASH is based on observation of the slightly different behavior of the reflexive and the subject gap. As will be seen in chapter 6, the fact that REFL and SLASH are FOOT features and obey the FOOT Feature Principle, while SUBCAT obeys the SUBCAT Feature Principle, will explain some of the ‘scrambledü sentences. Since the examples in this chapter are all ‘unscrambledü, almost identical results are obtained if we assume that the information on gaps is stored in SLASH instend of SUBCAT. However, we will see that the explanation of one example, namely (4.76), crucially relies on the fact that syntactic control of the subject gap is handled in terms of SUBCAT.
To be more precise, the set union operation is defined based on unification, rather than identity.
Z in the English translation shows that the referent of the reflexive is not determined intrasententially and is free.
There are some technical and methodological problems in the use of ‘designatedü variables in the semantics of reflexivization or unbounded dependencies, as in the earlier GPSG literature (e.g., Gazdar (1981, 1982a), Gunji (1983)). See Pollard (1983) and Pollard and Sag (1983) for discussion. The latter propose an alternative treatment of reflexivization in English. A similar treatment is presented for unbounded dependencies in Gazdar, Klein, Pullum, and Sag (1985).
The structure (4.51 b) is actually a substructure of (4.49c) with the C being P„0.
As is mentioned concerning (4.51b), syntactic control actually occurs at VI, which is a TVP node.
Note that as is mentioned in section 4.4.1, once the reflexive is bound, the REFL feature is no longer passed up to the mother by the FCR (4.41). Thus, a structure like the following is also prohibited, where syntactic control occurs only at V3, and hence the second occurrence of the reflexive is bound by the syntactic subject but the other remains free: Note that, although V3 satisfies the FCR, V2 does not, since the value of REFL is not empty, even though the PP[SBJ; +BRFL] in the SUBCAT is marked as the controller — i.e., its BRFL value is +. Thus, as shown in (4.55d), syntactic control occurs at both V2 and V3, binding both occurrences of zibun to the same antecedent. Note also that even if there is no syntactic control, the PP[SBJ]s in the REFL values of Pd. and V3 in (4.55c) unify by the FFP, since both unify with the PP[SBJ] in the REFL value of V2.
The S sign also shows the (BRFL +) specification on PP[SBJ]s. Two distinct symbols — and — are used to indicate that PP[SBJ]*s unify with one another, while PP[SBJ]$s unify with one another. These two sets of PP[SBJ]s do not necessarily unify with each other.
See also Miyara (1981) and Sugimoto (1982), which are both within the Montague-grammar framework. Interestingly, the former is interpretivist and the latter transformationalist. These recent formalizations, including Hasegawaüs and mine, depend on the existence of the VP node in such phrase structures as causatives and passives, an assumption that has not been very popular among some transformational grammarians of Japanese (cf. chapter 3).
Many recent analyses of reflexivization stress pragmatic explanations, which may or may not cover the same range of phenomena analyzed here. See Kuno (1978b, 1986a), for example, for such explanations.
Both the PP[SBJ] that is the controller of the missing subject and the PP[SBJ] that unifies with such a PP[SBJ] in the following trees are marked by a +. Although this corresponds to the * mark for reflexivization, there is no formal counterpart (i.e., (BRFL +)) for this mark; it is used for expository purposes only. The SLASH feature of a PP in the SUBCAT value of such nodes as V4 will be omitted for simplicity.
As mentioned above. a/ß in the following trees is an abbreviation for a [SLASH {ß}].
The syntactic control occurring at P[to] is ignored, since precisely the same kind of unification is repeated at V2.
If both reflexivization and subject control of the subject gap could occur at V„ we would get the following structure, which would correspond to the specification —, +, +, for RF at V2, RF at VI, and SC at VI, respectively. This structure, however, is not allowed due to the FCR (4.41), which prohibits the co-occurrence of a PP[SBJ; +BRFL] in the SUBCAT value and a non-null value of REFL. Note that both PP[to] and V2 violate this FCR.
As exhibited in (4.72a), it is not exceptional to have more than one gap in a sentence in Japanese. Hence, unlike English, the SLASH feature takes a set of categories, just like the SUBCAT feature. a[SLASH {ß, y}] will be abbreviated as a/{ß, y} in the following trees. In the tree and the logical translations below moo ‘moreü is omitted for simplicity.
The complement V subcategorizes only for an object. Although such a category looks strange, and in fact no lexical entry for it exists in Japanese, it will actually appear in a ‘scrambledü sentence; cf. chapter 6.
If two PPs unify, then the PFORM values must also be identical. Thus, strictly speaking, what has to be specified for the two PPs in the extended version of iw ‘sayü in (4.72d) is a somewhat weaker notion than unification, since it has the following lexical structure: (i) iw: V[SUBCAT {PP[SBJ], PP[OBJ; ni], PP[OBJ; to; SUBCAT {PP[OBJ; wo] Ill] where the PP[OBJ; ni] and the PP[OBJ; wo] must have identical feature values except for the PFORM value. Since the modification of the Control Principle based on a weaker notion (e.g., unify modulo PFORM) is straightforward, I will continue to use the term unify in the following.
As with a subject gap, an unbound object gap is translated into a (set of properties of a) free variable y, i.e., APP(v)
To be precise, such an interpretation is not available by syntactic control. It is, however, pragmatically possible, if the U (’Unspecified) in the English translation of (4.76) happens to be Ken according to the context. A quantifier can be used to exclude such pragmatic control, as discussed above. For example: (i) Naomi-no tomodati minna-ga Naomi-ni Tomio-ni kuruma-wo GEN friend all-NOM DAT DAT car-ACC kowasita-to iw-ase-ta. broke-COMP say-CA US-PAST All of Naomiüs friends made Naomi tell Tomio that U/she broke the car. In this case, the U cannot be pragmatically identified with the quantifier Naomi-no tomodati minna ‘all of Naomiüs friends,ü and hence the subject of the matrix sentence cannot control the missing subject pragmatically.
It might seem that the possibility remains of using the basic lexical specification of the causative as an input and obtaining the following as the output by a vacuous application of the set union operation in (4.45): However, if (i) is a possible output, then the PP[SBJ]s in the input must also unify, since the (vacuously) added PPISBJ]s unify with the already existing ones. Since such a lexical specification for sase does not exist, (i) is not a possible output, either.
We will see later (cf. chapter 6) that this difference between FOOT features and the SUBCAT feature explains some of the phenomena involving word-order variation.
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 1987 Springer Science+Business Media Dordrecht
About this chapter
Cite this chapter
Gunji, T. (1987). Control in Japanese. In: Japanese Phrase Structure Grammar. Studies in Natural Language and Linguistic Theory, vol 8. Springer, Dordrecht. https://doi.org/10.1007/978-94-015-7766-3_4
Download citation
DOI: https://doi.org/10.1007/978-94-015-7766-3_4
Publisher Name: Springer, Dordrecht
Print ISBN: 978-90-481-8239-8
Online ISBN: 978-94-015-7766-3
eBook Packages: Springer Book Archive