Es hat uns überrascht, das {sogar / auch nur} [der Hans]F da war. it has us ... scope theory and further extend the ana
Even, Only, and Negative Polarity in Japanese Kimiko Nakanishi University of Calgary 1. Scope vs. Lexical Theory of Even The English focus particle even is known to introduce a scalar presupposition (ScalarP) given in (1); Alan even saw [Bill]F presupposes that ‘that Bill saw Alan’ is the least-likely proposition among alternatives (LeastP). In contexts where negative polarity items (NPIs) such as any can occur, we observe ambiguity; I doubt that Alan even saw [Bill]F is ambiguous between LeastP and a reading where ‘that Alan saw Bill’ is the most-likely proposition (MostP). Two theories have been proposed to account for this ambiguity. One theory holds that the ambiguity is due to the scope interaction between even and a downward-entailing (DE) expression (e.g. doubt) (Kartunnen and Peters 1979). When even is under doubt, we obtain LeastP, while, when even takes scope over doubt, the likelihood scale gets reversed by the DE operator doubt, yielding MostP. The other theory holds that even is lexically ambiguous between a regular even with (1) and an NPI even with (2) (Rooth 1985). Cross-linguistic data have been provided in favor of the lexical theory; in the German example in (3), sogar ‘even’ allows LeastP only, while auch nur ‘also only’ allows MostP only, suggesting that sogar is a regular even and auch nur is an NPI even. Indeed, sogar, but not auch nur, is acceptable in positive contexts, as in (4). (1) (2) (3)
(4)
∀q[ [q∈C ∧ q≠p] → likelihood(q) > likelihood(p) ] [[even]]w (C)(p) w ∀q[ [q∈C ∧ q≠p] → likelihood(p) > likelihood(q) ] [[evenNPI]] (C)(p) war. Es hat uns überrascht, das {sogar / auch nur} [der Hans]F da it has us surprised that {even / also only} the Hans there was ‘It surprised us that even John was there.’ war. {Sogar / *auch nur} [der Hans]F da {even / also only} the Hans there was ‘Even Hans was there.’
2. Even in Japanese Just like (3), the Japanese examples in (5) seem to suggest that -mo ‘even’ and -demo ‘even’ are a regular even, while -dake-demo ‘even only’ is an NPI even licensed in DE contexts. Indeed, -mo/-demo, but not -dake-demo, is licensed in positive contexts. However, the distribution of these items in negative contexts casts doubt on this analysis. In negative contexts, only MostP is available in English, and auch nur, but not sogar, is licensed in German, as in (6). In contrast, in Japanese, -dake-demo, which behaves like an NPI even in (5), cannot be licensed in negative contexts, as in (7), while -mo/-demo can. Under the lexical theory, we may need to posit two types of NPI even, one licensed by negation (-mo/-demo) and one by other DE operators (-dake-demo). However, it is not clear why there are two types and why there are two forms for the first type, one of which (namely, -demo) shows an apparent morphological relation to the second type -dake-demo. In this paper, I account for the distribution of Japanese even under the scope theory and further extend the analysis to the data on Japanese NPIs. (5) a. John-ga [itiban muzukasii mondai]F{-mo/-demo/??-dake-demo} toita-to-wa odoroi-ta. John-NOM [most difficult question]{-even/-even/-only-even} solved-that-TOP was surprised ‘I was surprised that John even solved the most difficult question.’ LeastP
b. John-ga [itiban kantanna mondai]F{??-mo/??-demo/-dake-demo} toita-to-wa odoroi-ta.1 John-NOM [most easy question]{-even/-even/-only-even} solved-that-TOP was surprised ‘I was surprised that John even solved the easiest question.’ MostP (6) Niemand hat {*sogar / auch nur} [der Hans]F begruesst. nobody have {even / also only} the Hans greeted ‘Nobody even greeted Hans.’ MostP (7) John-wa [itiban kantanna mondai]F{-mo / -demo / *-dake-demo} tok-ana-katta. John-TOP most easy question{-even / -even / -only-even} solve-NEG-PAST ‘John didn’t even solve the easiest question.’ MostP 3. Even in Japanese under the Scope Theory The scope of negation in Japanese is known to be “narrow” (Kuno 1980); as in (8)a, negation tends to take scope over a quantificational element in a sentence. Crucially, as in (8)b, the focus particle -dake ‘only’ also takes scope over negation, unlike English only that takes the narrow scope. Then, it naturally follows that, without stipulating a covert movement, -mo/-demo ‘even’ take scope over negation, which yields MostP, as shown in (9). Moreover, the fact that -mo/-demo only allow LeastP, as in (5), suggests that they cannot move above other DE operators. That is, unlike English even, -mo/-demo do not undergo any covert movement at LF. (8) a. John-wa subete-no mondai-o question-ACC John-TOP all-GEN ‘John didn’t solve all the questions.’ b. John-wa [sono mondai]F-dake-o John-TOP that question-only-ACC ‘John didn’t solve only that question.’ (9)
tok-ana-katta. solve-NEG-PAST ∀>¬, ??¬>∀ tok-ana-katta. solve-NEG-PAST only>¬, ??¬>only
LF of (7): [ even C [ not [ John solved [the easiest question]F ] ] ] ScalarP: ‘That John didn’t solve the easiest question’ is the least likely proposition in C, i.e., ‘that John solved the easiest question’ is the most likely proposition.
I account for the distribution of -dake-demo by extending Guerzoni’s (2003) analysis on German auch nur with some modification. Guerzoni argues that a presupposition of auch contradicts that of nur, and that the contradiction can be revolved if auch can scope over a DE operator. In Japanese, -dake ‘only’ is assumed to introduce a presupposition in (10), which is inconsistent with ScalarP of -demo ‘even’ in (1). However, when there is a DE operator that -demo can take scope over, the conflict is resolved due to a scale-reversal property of a DE operator (LF: even>DE>only). The movement of -demo is well-motivated in that it is required to resolve a semantic conflict. Without the presence of -dake, which causes a semantic conflict, -demo cannot undergo any movement, as in (5). The compositional analysis of -dake-demo is capable of explaining why -dake-demo is unacceptable in positive and negative contexts. In positive contexts, there is no DE operator, hence there is no way to resolve a semantic conflict. In negative contexts, as in (8), -dake as well as -demo necessarily takes scope over negation, thus it is impossible to obtain LF: even>negation>only, i.e., the only configuration free of the semantic conflict. The current analysis predicts that -dake-demo is licensed precisely when there is a DE operator that -demo can scope over. The LF we obtain after resolving a semantic conflict 1
-mo is acceptable under the additive also reading.
yields MostP (-demo>DE), predicting that -dake-demo, whenever it is licensed, yields MostP only. As shown above, this prediction is borne out. Furthermore, the current analysis sheds light on differences between -mo and -demo: while their distribution seems to be exactly the same, they differ in that -demo, but not -mo, is compatible with -dake (-dake-demo vs. *-dake-mo). Unlike -demo, -mo can never undergo movement even with a semantic motivation. Then, the incompatibility of -mo and -dake (i.e., *-dake-mo, as opposed to -dake-demo) is explained by claiming that the semantic conflict between the two items can never be resolved. (10)
∀q[ [q∈C ∧ q≠p] → p >likely/insignificant… q ]
4. Further Predictions on NPIs The proposed analysis can further account for the distribution of Japanese NPIs consisting of the numeral one followed by -mo/-demo/-dake-demo. All three forms are unacceptable in positive contexts, while the distribution is more complex in DE contexts, as in (11) and (12). Following Lahiri (1998), the alternatives are introduced in terms of cardinality, as in {p: ∃n[p = λw.∃x[question(x) ∧ |x|=n ∧ solve(j,x,w)]]}. In positive contexts, the semantics of even says that John’s solving one question is the least-likely among the alternatives, which is inconsistent with the ordinary meaning of one in (13), i.e., one is the most-likely cardinality. If even can take scope over a DE operator, the inconsistency can be resolved (LF: even>DE>one) due to the scale-reversal property of DE operator. According to the current analysis, -mo can never move, hence one-mo is predicted to be unacceptable except for with negation. The prediction is borne out, as in (11) and (12). As for -dake-demo, it should be licensed in DE contexts other than negative ones; to resolve a semantic conflict, -demo moves above a DE operator, yielding LF: even>DE>only. Thus, one-dake-demo is licensed whenever the above LF obtains. Again, the prediction is borne out, as in (11) and (12). Regarding -demo, its distribution is the same as that of -dake-demo, which suggests that -demo in one-demo comes with a silent only. This is not implausible given that ScalarP of one in (13) and of -dake ‘only’ in (10) are always consistent. Then the distribution of -demo is explained exactly in the same way as that of -dake-demo. (11)
(12)
(13)
John-wa [iti-mon]F{-mo/*-demo/*-dake-demo} tok-anakat-ta. solve-NEG-PAST John-TOP one-CL{-even/-even/-only-even} ‘John didn’t solve even one question.’ John-ga [iti-mon]F{*-mo/-demo/-dake-demo} toita-to-wa solved-that-TOP John-NOM one-CL{-even/-even/-only-even} ‘I was surprised that John solved even one question.’ ∃x[ |x|=n ∧ solve(j,x,w) ] → ∃x[ |x|=1 ∧ solve(j,x,w) ]
odoroi-ta. was surprised
References Guerzoni, Elena. 2003. Why Even Ask? On the Pragmatics of Questions and the Semantics of Answers. Ph.D. dissertation, Massachusetts Institute of Technology. Kartunnen, Lauri, and Stanley Peters. 1979. Conventional implicature. In C. K. Oh and D. A. Dinneen eds., Syntax and Semantics 11: Presuppositions, 1-55. New York: Academic Press. Kuno, Susumu. 1980. The scope of question and negation in some verb-final languages. Proceedings of the Sixteenth Chicago Linguistic Society, 155-169. Lahiri, Utpal. 1998. Focus and negative polarity in Hindi. Natural Language Semantics 6, 57-123. Rooth, Mats. 1985. Association with Focus. Ph.D. dissertation, University of Massachusetts, Amherst.
