In this article, I argue that the phase edge in the C field shares features via Agree with an intermediate layer (FinP) and with a lower projection (⌺P), allowing it to determine the type of clause and its polarity. I adopt a feature-sharing relation of Agree that connects all of the polarity features present on heads (be they ⌺, Fin, and, in some cases, VFoc) to a polarity feature in Force, the relevant phase-edge position for clausal typing. This explains, among other things, why embedded clauses containing a polarity feature can only satisfy the selectional properties of a particular class of (matrix) verbs.(1) Q: Did Jason fall?A: Yes, he fell. AЈ: No, he didn't fall.(2) Q: Est-ce que Jason est tombé? A: Oui, il est tombé. AЈ: Non, il n'est pas tombé.(3) a. Je crois que oui/non. (Lit. I think that yes/no.) b. Peut-être que oui/non. (Lit. Maybe that yes/no.)(4) a. *I think that yes/no. (cf. I think so/not.) b. *Maybe that yes/no. (cf. Maybe so/not.)(5) a. Julie veut se marier, mais Anne, non. b. Julie n'aime pas les moules, mais les huîtres, oui/si.(6) a. ??Julie wants to get married, but Anne, no. (cf. . . . but Anne doesn't.) b. *Julie doesn't like mussels, but oysters, yes. (cf. . . . but she does oysters.)The article is organized as follows. In section 2, I give some background information about polarity particles in general, and English and French ones in particular. More specifically, I introduce several notions that are at the heart of the analysis defended here: relative and absolute polarity features (Farkas 2009), clausal typing (Cheng 1991), and the feature-sharing version of Agree argued for by Pesetsky and Torrego (2007). I conclude with a brief overview of how these notions can be used to account for the properties of English polarity particles and the ellipsis they sometimes license in echo assertions. In section 3, I first examine the distribution of polarity particles in elliptical embedded clauses and elliptical right conjuncts expressing contrastive polarity, and I argue that such particles lexicalize phase-edge clausal-typing features. I then address the question of why languages like French, Hungarian, Polish, and Romanian can effect clausal typing in embedded contexts via yes/no-type polarity particles but English cannot. This, I argue, follows from the inability of English-type languages to license embedded TP-ellipsis, leaving English with two options to express embedded polarity in a reduced form: so-pronominalization and VP-ellipsis. In section 4, I argue that the position of French polarity particles in the C field is the head of FinP, a projection that closes the C domain downward. I then show how the sharing version of Agree as applied to clausal-typing features explains how a French polarity particle located in Fin can inherit the ( positive or negative) value of Laka's (1990) ⌺ and ''pass it on'' to Force, allowing a matrix verb to ''see'' whether the clause it has merged with satisfies its selectional properties. In section 5, I consider special uses of French polarity particle...