Particle-Stranding Ellipsis in Japanese, Phase Theory, and the Privilege of the Root

This squib analyzes a type of ellipsis in Japanese, which I dub particle-stranding ellipsis (PSE). This elliptic pattern, first documented by Hattori (1949, 1960), has been discussed in subsequent work (Arita 2009, Hayashi 2001, Sato 2008, Sato and Ginsburg 2006, 2007, Vance 1993, Yoshida 2004) but has not received due attention in the literature. PSE is used as a truncated reply in colloquial dialogues, as shown in (1).¹ In this dialogue, where Speaker A’s question introduces Tanaka as the topic of conversation, Speaker B’s response begins with the noncontrastive topic marker -wa.

Rizzi (2005b) proposes within phase theory (Chomsky 2000, 2001, 2004) that the edge of the root may remain unpronounced while still being accessible to semantic interpretation, and he applies this analysis (“the privilege of the root” (PoR)) to topic drop in German (Ross 1982). I propose that PSE instantiates the PoR phenomenon; it arises when the topmost phase head is spelled out at the end of the phase it defines, together with its complement, for phonetic interpretation at PF, while the entire structure is transferred to LF for semantic interpretation.

Yoshida (2004) observes that PSE has three distributional properties. First, it can apply only to a sentence-initial topic, as the contrast between (2a) and (2b–c) shows.

Second, PSE is a root phenomenon, as shown in (3a). (3a) becomes grammatical with Taroo-wa in the embedded CP, as in (3b). This minimal pair thus shows that (3a) is judged ungrammatical not because there is an embedded topic but because the embedded topic is null.

Finally, PSE can occur only once in a clause, as shown by the contrast between (4a) and (4b).

PSE is different from so-called argument ellipsis (AE) in Japanese (Kim 1999, Oku 1998, Saito 2004, 2007, Takahashi 2008) because AE exhibits none of the properties that characterize PSE. First, AE can target a non-sentence-initial element, as in (5b–c).

Notice that the surface order in (5a–c) itself does not conclusively tell whether AE occurs in the positions indicated by Ø, because Japanese is a free-word-order language. Unless evidence to the contrary is presented, I simply assume that the ellipsis of an NP occurs in the surface positions it would occupy were it pronounced. Indeed, the variants of (5a–c) with an overt pronoun in the gap are grammatical, as shown in (6a–c).

Second, AE is not a root phenomenon; it can occur in an embedded clause, as in (7).

Finally, AE can occur more than once in a clause, as illustrated in (8).

The distributional differences between PSE and AE therefore suggest that the former needs a different treatment from the latter (e.g., LF copy in Kim’s (1999) and Oku’s (1998) theory).

Within the standard version of current phase theory (Chomsky 2000, 2001, 2004), Spell-Out applies to the complement of a phase head. Rizzi (2005b) argues that the PoR phenomenon obtains when an element in the specifier of the topmost phase escapes Spell-Out. Consider how this theory applies to German topic drop, illustrated in (9b–c).

In (9b–c), a topic moves overtly into Spec,Top. “Topic drop” arises when the phase head Top spells out its complement, leaving the topic inaccessible for pronunciation.

As pointed out by Yoshida (2004), this construction exhibits the same patterns as PSE. First, it can only occur in the sentence-initial topic position, as illustrated in (10a–b).

Second, as observed by Cardinaletti (1990) and Rizzi (1994, 2005a), German topic drop is a root phenomenon. A null topic cannot occur in an embedded clause even when it occupies the clause-initial topic position. This point is illustrated in (11a–b).

Finally, German topic drop can occur only once in the root clause, as shown in (12).

I propose within a minimum modification of the standard phase theory that PSE obtains when the phase head Top may optionally be sent to PF at the end of the phase it defines, together with its complement TP. This analysis is shown in (13) for Speaker B’s answer in (1).

In (13), the noncontrastive topic marker -wa occupies the Top phase head (see Kayne 1994 and Whitman 1997 for evidence for the clause particle analysis of -wa). The size of the syntactic object sent to the interfaces is different in (13). The Top head and its complement are spelled out to PF for phonetic realization whereas the entire structure including the specifier of the phase head is transferred to LF for semantic interpretation.²

The three properties of PSE follow straightforwardly. If the TopP in the root clause is preceded by any other element in the specifier of a higher discourse-functional head in the left periphery (e.g., Foc; see Rizzi 1997), the noninitial TopP is inevitably contained in a higher phase. Hence, the specifier of Top is spelled out in the higher phase. This analysis correctly blocks the derivation for utterances like Speaker B’s shown in (14) where TopBBis first spelled out, leaving the specifier of Top unpronounced, followed by the spell-out of the PP in the specifier of a higher functional head (such as Foc).

Similarly, PSE cannot occur in an embedded context because then the specifier of TopP would be spelled out in the next higher phase and would necessarily be phonetically realized. For the same reason, there cannot be more than one PSE in a single clause because the second instance would be within a higher phase.

Sato and Ginsburg (2006, 2007) observe that the complement NP of a postposition may also be elided, as shown in (15).

Following Abels (2003) and Van Riemsdijk (1978), suppose that PPs are phases. Since phases are associated with the EPP feature, the NP complement of the P is moved to Spec,P. Then, the string Ø -kara-wa results when Spell-Out applies to the P head (i.e., kara ‘from’) as well as to the Top head plus its complement, as represented in (16).³

Note that this analysis also correctly predicts that kara ‘from’ may also be elided in Speaker B’s utterance in (15) because the spell-out of the phase head itself is optional.

This squib has argued that PSE exemplifies the PoR phenomenon. It has proposed that stranding of the noncontrastive topic marker -wa arises when the Top phase head is optionally spelled out to PF, together with its TP complement, for phonetic interpretation at PF whereas the entire syntactic structure including the topic is transferred to LF for semantic interpretation. This analysis lends support to a nonsimultaneous transfer model of grammar (e.g., Marušič 2005) according to which linguistic interfaces target different parts of the derivation at a phase level.