Against Subsegmental Glides

This squib claims that subsegmental glides should be eliminated as licit representations in phonology. Section 1 asks whether segmental glides are necessary, a question that is answered affirmatively in section 2. Sections 3 and 4 consider the evidence for subsegmental glides and conclude that there is none. Section 5 sumarizes the conclusions and compares the predictions made by subsegmental and segmental representations of glides.

The illustrative material presented here is drawn primarily from Ukrainian and Dutch because these languages show directionality effects and thus bear in a crucial way on the potential evidence for subsegmental glides. In addition, the choice of Ukrainian is rewarding in one more way: the data have never been discussed in the generative literature to date.

1 Representations

The idea that glides can be subsegmental originated with McCarthy and Prince (1986) and stems from their effort to reconcile the Obligatory Contour Principle (OCP) with the moraic skeleton (see also Hayes 1989, Rosenthall 1994). Subsegmental glides do not have a Root node of their own. Rather, they share a Root node with the high vowels that spawn them and, consequently, they do not violate the OCP. This is illustrated in (1), which shows instances of glide insertion spawned by /i/. The glide is inserted either to the left of i (1a) or to the right of i (1b), depending on the configuration of [i] and another vowel (here, taken to be an [a]). Dots denote syllable boundaries.

(1)

1. a. Leftward spreading: /ai/ → [a.ji]

2. b. Rightward spreading: /ia/ → [i.ja]

Since McCarthy and Prince 1986, the subsegmental representation of glides has been assumed either tacitly or openly by many researchers working in the moraic framework. For example, McCarthy and Prince (1995) use it in their analysis of Malay without even considering the alternative that glides might be segmental-that is, that they might have their own Root node, as shown in (2), which is based on Hayes 1989 and which clearly violates the OCP.

(2)

1. a. Leftward spreading: /ai/ → [a.ji]

2. b. Rightward spreading: /ia/ → [i.ja]

However, it is not clear that segmental glides actually need to violate the OCP because the alternative to (2) is to represent [ji] and [ij] as two Root nodes with shared features, as in (3).¹

(3)

1. a. Leftward spreading: /ai/ → [a.ji]

2. b. Rightward spreading: /ia/ → [i.ja]

Regardless of whether (3) does or does not violate the OCP, the issue has largely lost its force in Optimality Theory (Prince and Smolensky 1993, McCarthy and Prince 1995), because constraints are violable and the OCP should be no exception.

While the OCP as an argument for subsegmental glides may have disappeared, the problem of how to represent glides is still with us. As mentioned earlier, McCarthy and Prince (1995) assume subsegmental glides without discussion. Further, Rosenthall (1994) argues that sub-segmental glides are the default representation because they have less structure than segmental glides and are therefore favored by the OT *STRUC constraint. However, Rosenthall does not present empirical arguments for subsegmental glides. This gap is filled by Keer (1995, 1996), whose arguments I consider in section 3 and, to some extent, in section 4. If subsegmental glides are the default case, the question is whether segmental glides are necessary at all. The answer is affirmative, as I argue below by looking at glide insertion in Ukrainian and Dutch.

2 Segmental Glides

Ukrainian and Dutch present two different arguments for segmental glides. First, glides must be able to fall within the purview of ALIGN-L(EFT) (Ukrainian). Second, glides may come from partial spreading; that is, they are not copies of the vowels that spawn them (Dutch).

In Ukrainian /i/ induces j-insertion in order to provide an onset.² However, j-insertion operates only word-medially (4a) and not word-initially (4b). These two generalizations can cooccur in the same word (4c). Glide insertion is not sensitive to whether the syllable is stressed (shown by an accent here) or unstressed.

(4)

1. a. stójik³ 'Stoic', kokajín⁴ 'cocaine'

2. b. Iván 'Ivan', íra 'Irene'

3. c. Izrájil 'Israel', intujitývnyj 'intuitive...