Nominal Ellipsis and the Interpretation of Gender in Greek Yasutada Sudo and Giorgos Spathas (University College London and University of Stuttgart) This paper provides a new analysis of nominal ellipsis involving gender mismatches in Greek, discussed by Merchant (2014; M). M shows that Greek masculine-feminine pairs of nouns form three classes with respect to the behavior under nominal ellipsis (see also Bobalijk & Zocca 2011 for Brazilian Portuguese). We argue with novel data that (i) gender in the nominal domain is in some cases part of the assertoric component, not just presupposed, and that (ii) nominal ellipsis with gender mismatches is possible in both predicative and argument positions. These observations are problematic for M’s analysis. We claim instead that nominal ellipsis in Greek always involves PF-deletion (contra M), and that the principle of Maximize Presupposition should be made sensitive to overt material only. Three noun classes: M observes that there are three classes of masculine-feminine noun pairs that differ in whether nominal ellipsis involving gender mismatches is possible (ellipsis is always licensed with matching gender): With Class I nouns like adherfos-adherfi ‘brothersister’ ellipsis with different genders is impossible, unlike with nouns in Class II like jatros ‘doctor’. Class III nouns like dhaskalos-dhaskala ‘teacher’ are compatible with nominal ellipsis when the antecedent is masculine, but not when the antecedent is feminine, (1). (1) a. O Petros ine kalos dhaskalos, ala i Maria ine mia kakia e. the Petros is good.m teacher.m but the Maria is a.f bad.f e ‘Peter is a good teacher, but Mary is a bad one.’ b. * I Maria ine kali dhaskala, ala o Petros ine enas kakos e. the Maria is good.f teacher.f but the Petros is a.m bad.m e M’s analysis is based on two assumptions: (i) Gender features in the nominal domain have presuppositional meanings. Nouns in different classes differ in whether they carry such features, (2)-(4). (ii) Next to a PF-deletion strategy, Greek nominal ellipsis can also be the result of a null proform eN. It is this second strategy that can license ellipsis in cases of gender mismatch. By assumption eN refers to a contextually salient property denoted by a noun in the discourse. In (1a), eN is resolved to the antecedent dhaskalos. Since it has no presupposition of its own, (4a), resolving eN to it will not cause a semantic problem, even if the subject of the second sentence is feminine. By contrast, in (1b), eN is resolved to dhaskala, which has a lexically specified gender, (4b), and causes a semantic clash with the masculine subject. Since Class I nouns are always specified for gender, semantic clashes arise in both directions. Class II nouns, which are never specified, never lead to such clashes. (2) [[adherfos]] = λxe: male(x). sibling(x) (4) a. [[dhaskalos]] = λxe: teacher(x) [[adherfi]] = λxe: female(x). sibling(x) b. [[dhaskala]] =λxe: female(x). teacher(x) (3) [[jatros]] = λxe: doctor(x) Some genders are asserted. We provide independent evidence that some nouns assert, as well as presuppose the gender inferences, contrary to M. It is known (Heim 1994, 2005, 2008, a.o.) that φ-features on bound pronouns that receive presuppositional interpretations (like gender) have no effect in the focus alternatives in examples like Only Mary did her homework, i.e. the sentence entails that both male and female alternatives to Mary didn’t do their homework. In such contexts, Class III masculine nouns behave like possessive pronouns, e.g. (6a) entails that Maria is not a teacher. On the other hand, (6b) with a Class III feminine noun does not entail that Petros is not a teacher. Class I nouns behave similarly to (6b), Class II nouns similarly to (6a). We, thus, propose the semantics of nouns in (7)-(9). (6) a. Mono o Petros ine dhaskalos. b. Mono i Maria ine dhaskala. only the Petros is teacher.m only the Maria is teacher.f ‘Only Peter is a teacher.’ ‘Only Maria is a female teacher.’ (7) [[adherfos]] = λxe: male(x). male(x) & sibling(x) (8) [[jatros]] = λxe: doctor(x) [[adherfi]] = λxe: female(x). female(x) & sibling(x)

(9) [[dhaskalos]] = λxe. teacher(x) [[dhaskala]] = λxe: female(x). female(x) & teacher(x) As is well known, the identity condition on ellipsis generally ignores presuppositional φfeatures like gender (cf. Fiengo & May 1994; e.g. Mary did her homework. John didn’t do his homework.), but not, of course, differences in assertive content (cf. *John’s brother lives in London. Bill’s sister lives in Stuttgart.). Then, given (7)–(9), the behavior of Class I and Class II nouns under ellipsis with a gender mismatch is expected. The asymmetry in the case of Class III nouns in (1) can also be explained assuming a (syntactic and/or semantic) total identity condition on ellipsis. In this case the structure of (1) is as in (10). (10b) is simply pragmatically out, because its gender presupposition won’t be satisfied. No problem arises in (10a), since dhaskalos is not specified for gender. (We assume that gender features on the determiner and the adjective are uninterpretable agreement features licensed by a semantically interpreted gender feature outside the DP, as in Sauerland 2003.) (10) a. O Petros ine kalos dhaskalos, ala i Maria ine mia kakia dhaskalos. the Petros is good.m teacher.m but the Maria is a.f bad.f teacher.m ‘Peter is a good teacher, but Mary is a bad one.’ b. * I Maria ine kali dhaskala, ala o Petros ine enas kakos dhaskala. the Maria is good.f teacher.f but the Petros is a.m bad.m teacher.f Maximize Presupposition! However, the overt version of (10a) is ungrammatical, (11), although according to our analysis, there should not be a semantic clash. (11) *O Petros ine kalos dhaskalos, ala i Maria ine mia kakia dhaskalos. The problem is the same as in simple predicative cases like (12). If dhaskalos is not specified for gender, what causes the infelicity of (11) and (12)? To handle similar cases, Heim (2008), a.o., invokes the principle of Maximize Presupposition (14) (see also Percus 2006, Sauerland 2008). We propose two refinements of MP to deal with (11) and (12), (15). Firstly, to explain the infelicity of (12) with the alternative in (14), we drop the first clause of MP, a move Spector & Sudo (2014) propose on independent grounds. Furthermore to explain the difference between (11) and (10a), we further propose that MP only looks at overt material. This is independently motivated by examples involving a vs. the and both vs. all e.g. (16). (12) *I Maria ine dhaskalos. (13) I Maria ine dhaskala. the Maria is teacher.m the Maria is teacher.f ‘Maria is a teacher.’ (14) Maximize Presupposition! (MP) (Heim 1991) Sentence S is infelicitous in context c if there is an alternative S’ such that a. S and S’ contextually entail each other in c; b. S’ has a stronger presupposition than S; and c. the presupposition of S’ is satisfied in c. (15) Maximize Presupposition!* (MP*) Sentence S is infelicitous in context c if there is an alternative S’ such that a. the presuppositions triggered by overt items in S’ are stronger than those triggered by overt items in S; and b. the presuppositions of S’ are satisfied in c. (16) John has three sisters. Bill has two sisters. John likes all of his sisters. a. Bill likes {both / *all} of his sisters too. b. Bill does like all of his sisters too. Argument ellipsis. M also argues nominal ellipses with gender mismatches are never possible in argument position. We argue that his data with argument ellipsis involve several independent confounds (regarding contrastive topic), and show that once these confounds are removed, e.g. (17), argument ellipsis exhibits exactly the same pattern as predicative ellipsis. (17) O Petros episkefhike enan dhaskalo sti Veria, ke mia e stin Katerini. the Petros visited a.m teacher.m in.the Veria, and one.f e in.the Katerini ‘Peter visited a male teacher in Veria, and a female teacher in Katerini.’