SEPHARDIC SCANSION AND PHONOLOGICAL THEORY.

• Author: Hoberman, Robert D.
• Position: Hebrew poetry in medieval Spain

1. A good theory should be able to predict facts that can then be tested empirically. That is what theories are for. While this is standard practice in the natural sciences, in philology we rarely get a clear-cut instance. However, in the case of the metrical pattern of Sephardic Hebrew poetry we have a very strong candidate. Manuals of Hebrew poetics have often mentioned that the conjunction u- 'and' scans on some occasions as long and on others as short. Joseph Malone (1983) discovered the key regularity in this variation. Given the facts that Malone had discovered, the theory of the phonological aspects of poetry which Manaster Ramer has been developing led us to hypothesize that the actual pronunciation of Hebrew by the medieval Sephardic poets was different in a specific detail from what has been commonly assumed. When we subsequently looked for evidence for medieval Hebrew pronunciation we learned of documentary evidence (edited and interpreted by Geoffrey Khan and others) which confirmed exactly the detail we had predicted on the basis of a simple, empirically predictive theory.

   It is traditional for proponents of different linguistic theories, and especially theories of phonology, to seek validation by explaining the operation of systems of versification. Indeed, it seems as if for over a century every new phonological theory has been associated with claims that a level (or levels) of representation posited by that theory functions uniquely in the poetic systems of all languages: it is that level which the rules of versification "look at" to determine whether and how any particular sound or stretch rhymes, alliterates, scans, etc. In other words, metrical equivalence, to the extent that it depends on phonology at all, is claimed to be defined at different levels of representation depending on the phonological theory, and hence facts of metrical equivalence are used to distinguish between phonological theories which may otherwise be hard to choose among. Generative phonology, since its inception, has yielded a particularly rich crop of such arguments (such as Zeps 1963; Schane 1968; Kiparsky 1968, 1972, 1973; Anderson 1973; and others). There have been rather fewer of these lately, perhaps because it is now widely accepted (as shown by such survey articles as O'Connor 1982: 155-56 and Hayes 1988: 228-29) that there do exist poetic systems which define metrical equivalence in terms of the kinds of representations postulated by generative phonology.

   On the other hand, Manaster Ramer (1981, 1994, 1995, and forthcoming) has sought to refute these arguments and to motivate a universal constraint on phonological theory to the effect that no rules of versification may have access to any level of representation deeper or more abstract than the classical phonemic. This is in contradistinction to generative theory, which proliferates abstract levels. In recent years, related but weaker claims have been advocated by Mohanan (1986: 199) and by Kiparsky in unpublished work (personal communication; see Manaster Ramer 1994:317-18 for discussion). In this paper, we continue the tedious but necessary job of refuting one-by-one every example ever cited where distinctions made at levels of representation deeper or more abstract than the classical phonemic have been claimed to be reflected in the facts of versification, and to offer alternative analyses. This paper is thus part of an ongoing effort to demonstrate the validity of a view of versification that is both much older and much more constrained than that of many generative phonologists. The focus this time is on Malone's discussion (1983 and 1996: 121-22) of the meter of poems composed in medieval Spain in a variety of Hebrew which we may call Sephardic and in a metrical system borrowed from Arabic.

   An earlier version of this article challenged Malone's abstract generative analysis of Sephardic metrics by showing four different conceivable treatments of the problem, none of which refers to levels of representation deeper than classical phonemic. Two of the treatments adopted Malone's assumptions as to how Hebrew was pronounced by the poets, while the other two hypothesized that the pronunciation differed from those assumptions in certain details in such a way that the metric pattern fit our theoretical approach more directly. All four of these alternative treatments were conceived entirely under the guidance of our theoretical point of view, without examining evidence other than that cited by Malone. After that version of the paper had been accepted for publication in this journal we began to look into the evidence which exists regarding the medieval pronunciation of Hebrew. A new body of research on this topic has appeared recently, based on transcriptions of Hebrew into other alphabets, mainly Arabic, and on highly detailed contemporaneous descriptions of the pronunciation. We were gratified to learn from this body of research that in crucial detail Hebrew was pronounced exactly as we had proposed hypothetically on the basis of a theoretical exercise.

2. We begin by presenting the problem and summarizing Malone's generative solution for it. Sephardic Hebrew as described by Malone had two series of vowels: full, which Malone transcribes as i, e, [UNKNOWN TEXT OMITTED], a, a, [UNKNOWN TEXT OMITTED], o, and u, and reduced, [UNKNOWN TEXT OMITTED]. The full and reduced vowels were written differently in the traditional writing system and are assumed to have been pronounced differently. (We will show in section 4 that the reality was somewhat more complex.) In the metrical system, the full vowels scanned as long and the reduced vowels as short. While in almost all cases the long/short distinction in the meter refers to the surface phonology, as reflected directly in the orthography, Malone identifies one allomorph of one morpheme where, he argues, a nonsurface distinction is relevant to the meter.

   This is the u- allomorph of the morpheme meaning 'and', whose underlying form Malone takes to be/wa-/. This allomorph is written as a full vowel but counts in the meter sometimes as long and sometimes as short, depending on the precise environment in which it occurs. The u- allomorph occurs in two environments: (a) in an open syllable before a labial consonant (which is itself followed by a vowel), e.g., [UNKNOWN TEXT OMITTED] 'and (it) died', u-vallayil 'and at night'; and (b) in a closed syllable, before a cluster of two consonants (the first of which is neither a glottal nor a pharyngeal), in which case it scans as long, e.g., [UNKNOWN TEXT OMITTED] 'and see!', [UNKNOWN TEXT OMITTED] 'and drink!', u- v anane 'and by (the) clouds (of)'. In these two environments u- scans differently. When occurring in an open syllable (a) it scans as short; when occurring in a closed syllable (b) it scans as long.

   In order to account for the difference in scansion, Malone argues that u-, although it is derived from the same underlying form, goes through different stages of derivation in the two cases. According to his analysis, underlying /wa/ goes, in most cases, either to [UNKNOWN TEXT OMITTED] via a rule of Reduction (in open syllables) or to [UNKNOWN TEXT OMITTED] via a rule he calls Checked Midding (in closed syllables). Each of these intermediate forms is subject to various...