¹ For a brief discussion and summary of the facts see A. Meillet and J. Vendryès, Traité de grammaire comparée des langues classiques ³ 57–8 (Paris, 1960).

² With one possible exception. The Greek reflex of the Indo-European root *g^whedh appears in Greek in the first aorist form théssasthai ‘to pray for’, in the nominal form póthos ‘yearning’, and in the derived verb pothéō ‘I yearn for’. However, because of the defective nature of this root in Greek, it is possible to assume that at the time that the restriction in the domain of application of Grassmann's law in Greek came about, the initial stop of théssasthai had already been rephonemicized as /th/, and hence as a dental stop retained its inherited aspiration.

³ The statement of the phonological rules in this paper conforms to the notational conventions established by Halle, Chomsky, and other generative phonologists. In particular, the subscript one appearing in Rule 1 is to be read ‘one or more segments’. Triliteral abbreviations for the Jakobson-Halle features are used throughout, ‘cnt’ means ‘continuant’.

In connection with Grassmann's law, I am not concerned here whether the law is an independent innovation in Greek and in Indic, or whether it antedates the split of the two families. The former assumption is made by the neogrammarians, and a typical argument is given by Leonard Bloomfield, Language 350–1 (New York, 1933). Recently, in ‘Analogy and sound change’, a paper read at the 1964 annual meeting of the Linguistic Society of America, Paul Kiparsky has challenged the neogrammarian assumption within the framework of generative phonology.

⁴ Table 1 summarizes Rules 1 and 2 of Winfred P. Lehmann, Proto-Indo-European phonology 17 (Austin, 1955). Permitted combinations of initial and final stops are illustrated by ‘typical’ roots, and prohibited combinations are indicated by a dash.

⁵ If we disallow the use of the term rephonemicization for this change, it would be useful to have a technical term to cover this particular change and others like it, for example the Latin development *p…k^w→k^w…k^w. One good candidate is the term respecification, since these changes involve respecifying the values of particular redundant features in particular contexts.

⁶ The use of the variable α in the statement of this rule conforms to the conditions on the use of variables in phonological rules as stated by Noam Chomsky, Aspects of the theory of syntax 175–6 (Cambridge, Mass., 1965). It will be noted that the effect of Rule 2 is to prevent the occurrence of a root-final nongrave aspirated stop following a root-initial aspirated stop, since an Indo-European root does not begin and end with the same phoneme (Lehmann 17). As is noted in Meillet and Vendryès 58, medial /th/ in Greek is subject to a special rule which deaspirates it when another aspirate precedes. The effect of Rule 2 together with the rule that prevents a root from beginning and ending with the same phoneme is to prevent this special rule involving /th/ from ever applying within a root.

⁷ Hesychius, however, cites a form sunthúksō, which, although more recent in root vocalism and in voice, has the expected initial /th/. Calvert Watkins, to whom I am endebted for the Hesychius citation, has also pointed out to me that an initial dental stop regularly retains its aspiration in inherited diaspirate roots only if it is followed by /r/, and in two cases by /a/. Those cases are the roots thakh and thaph, appearing in the words takhús ‘swift’, thássōn ‘swifter’ and in tháptō ‘I bury’, táphos ‘grave’. If we incorporate these restrictions into the statement of Rule 2, then the form teukh is no longer exceptional.