1. Core claims: This paper argues that many of the cases of allomorphy analysed in the literature via conditioning environments for morpheme insertion should be subsumed under the syntactic operation Agree. This opens the way to a restrictive theory of morphology in which adjacency is the only conditioning environment for the insertion of the phonology of lexical items (Vocabulary Insertion (VI) in the sense of Halle & Marantz 1993). It also allows us to define adjacency in the strictest possible fashion as simple sisterhood within X? Together with the theory of Agree developed in Chomsky 1999, 2000, this predicts: If Agree only applies between functional heads, it cannot be the cause of root allomorphy (suppletion). We therefore predict that suppletion is solely conditioned under adjacency. In cases where it appears that two or more factors condition suppletion, it follows that all but one of them is the product of Agree, and therefore subject to the constraints in (iii) and (iv). Affixal allomorphy may be long-distance, but cannot cross a phase boundary. Affixal allomorphy is subject to syntactic intervention effects.

2. Mechanisms for Allomorphy: Standardly, allomorphy and suppletion are assumed to arise because special phonological forms are inserted in certain morphosyntactically defined contexts. Contexts are unrestricted enough that they can include non-local information, a property that has been used to analyse cases of apparent long-distance conditioning of allomorphy. We propose that contexts should only involve adjacent (sister) elements within an X?(which may be derived via head-movement, or 'lowering' (Morphological Merger in the sense of Marantz 1988)). This predicts that long-distance allomorphy should not exist.

2.1 Root Allomorphy: In part this prediction is true. Long distance root allomorphy is unattested, as far as we are aware. Typical cases of root allomorphy involve cases where the root and the conditioning environment are sisters within X? shown by the fact that intervening morphemes block suppletion. For example, some Scottish Gaelic nouns supplete for number, but this is blocked if a diminutive suffix is added: bean ~ mnaoi "wife" ~ "wives", beanag ~ beanagan / *mnaoiag "little wife" ~ "little wives").

lie several: k'ul / neg: kopgau lay several: k'op / neg: k'ugu How do you say: [I:them]-lay-distributive-neg = de-k'op/k'up/k'u-gu,-mau I camped in several places I didn't go and camp in the woods: adom haun be-k'op/k'up/k'u-ho'`-gu

A similar case can be made for Kiowa, where verbs supplete for their features of their object (t'al "sever one", tha' ` "sever several").

We argue that this is because the verb is a sister, after head-movement, to little v, which probes features of the object (hence object agreement). When an evidential affix intervenes between the theme vowel and the verb root, no suppletion for object takes place. However, there do seem to be cases where roots supplete depending on properties of heads further away than the adjacent head. For example, in Georgian, the verb qv, "take", suppletes in a 4-way paradigm for number of its object and tense. We argue that, in this case, little v is the goal of a probe on T (as evidenced by allomorphy of the theme vowel for T), and it is adjacency to the valued little v that provides the insertion context for the verb root. A similar case is found in Gaelic, where certain verbs supplete for tense and for conjunct morphology (verbs are conjunct when C has certain properties). This gives rise to a 4 way distinction once more, but we show that the conjunct relation is one where C values a feature of T, and the appropriate suppletive form of V is only inserted when V has raised to T. When the V stays in VP (as in cases where T is filled), no suppletion of V for conjunct morphology takes place.

2. 2 Affixal Allomorphy: Bobaljik 2000 discusses some interesting cases of long-distance allomorphy in Itelmen, and shows that the form of the class suffix on certain verbs is dependent on the featural composition of a higher object suffix, and a yet higher subject prefix. He argues that adjacency cannot be a sufficient condition to capture all the data and admits insertion contexts for the allomorph which access non-local information in the tree. We argue that the morphological component should not have access to such information, and that this allomorphy is actually a result of two syntactic probe-goal relations between higher functional heads and little v. Similar long-distance allomorphy can be found in Kiowa (Tanoan), where the transitivity of the verb (marked, we assume on little v) correlates with particular forms of both future and negative suffixes. In the case of the future, intervening morphemes do not disrupt this relationship, while they do with the negative. We argue that the v-neg relationship is allomorphy conditioned under adjacency, while the v-future relationship is simply Agree. Since allomorphy may take place under Agree, we predict that we may find intervention effects on allomorphic relations. Georgian bears this prediction out. We show that the allomorphy patterns of the agreement prefix are conditioned by whether there is a closer object in the syntax which is probed by certain (marked) features of little v. In such cases, the prefix agrees with the features of the object rather than the subject. Finally, we predict that allomorphy relationships which arise from Agree should only be possible when they access the edge of a lower phase. This rules out such cases as verbs which dispay allomorphy (say transitivity marking, or object agreement) for the tense of their finite complement clause, or for properties of a lower subject of a finite complement clause. These predictions appear to be bourne out cross-linguistically.

3 Theoretical Ramifications: In addition to these empirical predictions, the system has the theoretical advantage that the morphological component's access to the syntax can be restricted to X?elements only, and that no relation of government need be defined for VI (contra Halle & Marantz 1993). A highly constrained picture of the syntax-morphology interface emerges.