Modern Persian verb stems revisited.

Henderson, Michael M.T.

1. More than a decade and a half ago I described the verb morphology of modern Persian as a six-slot string of constituents, each slot rewritten as a pair of features or as a phonological matrix(1). I The second section of that article listed the five large classes of verb stems, divided according to the differences between the present and

(1) invariants, such as mandan `stay' xordan `eat'

(23 verbs(2))

(2) consonantal alternations, such as bastan/band(3)

`tie' (70 verbs)

(3) vocalic alternations, such as burdan/bar `carry'

(15 verbs)

(4) augmentative stems, in which the past stem is one

or two segments longer than the present stem,

such as danistan/dan `know' or nihuftan/nih

`wear' (18 verbs, plus the productive i-augment

class)

(5) exceptional or suppletive verbs in which no responsible

suggestion can be made that the alternations

are rule-governed, such as bu/bas/ast

`be' and dadan/bin `see' (18 verbs).

Other members of class (5) may share one or more alternations with members of another class, such as zududan/zida `rub off', which shares a process with asudan/asa `rest' but has another, unique vowel alternation earlier in the stem.

A linguist writing rules to describe linguistic phenomena such as these is trying to account for the knowledge that native speakers of the language have in their minds, and to do so in a manner which is consistent with a plausible view of a child's ability to acquire language. It is a plausible assumption, for instance, that confronted with a reasonably large set of examples of a particular phenomenon, a child will construct a rule which can be applied to as-yet-unheard forms. We know that this happens, because we have examples of children's overgeneralizations: English speaking children apply the regular past tense formation rule (add -d, devoiced to t after voiceless consonants, and insert [unknown character] first if the verb stem ends in t or d) to verbs which have irregular pasts: the English-speaking world is full of children producing forms like goed, comed, and sended. Effors like this tell us that our rules are in fact descriptive of rule-governed behavior. But we also write rules, such as the ten I listed in "MPVM," which look just like the kinds of rules a child might construct to account for a large body of data, but which in fact account for only one or two forms. These rules are just like the kind we could write to account for the many strong, verb alternations of English which have remained in the language, such as [unknown character] [left arrow] ey for the come/lcame alternation. It looks like a rule, and it could be written with distinctive features, but a rule is commonly accepted to mean "a description of a regularly occurring phenomenon." Just as it takes at least three aligned trees to make a row, it certainly takes more than one or even two examples to be a rule-governed occurrence.

In "MPVM" (p. 384) I suggested that "a generative grammar is written in terms of rules, but ... this does not imply that everything the grammar describes is rule-governed behavior." Nevertheless, the vexing question remains of how to write a description which adequately describes the facts, as "MPVM" did, while at the same time plausibly reflecting the kinds of knowledge a Persian-acquiring child might have, which "MPVM" did not.(4) In the time since the publication of that article, a new approach to the description of morphophonology has been developed, one which helps to deal with the welter of historical relics which makes up the Persian verb stem system, and it is to the application of this newer model that I shall now turn.

2. Lexical phonology. It has long been assumed that while the regularly occurring, predictable parts of language should be described by rule, the unpredictable parts belong in lists. Lists hold the forms that a learner must memorize, consciously or subconsciously, since they are not susceptible to rule formation. The list that linguists assume exists in everyone's mind is the lexicon. This is where all the morphemes go, at least those with semantic content,(5) and where the particular characteristics of each morpheme are listed: its form, grammatical category, semantic features (such as the animacy or inanimacy of a noun, or the requirement that a verb's object be animate, as is the case for frighten), and any other unpredictable facts about the morpheme. Beyond these facts, it is now considered wise to put some processes in the lexicon, processes which look for all the world like phonological rules, but which apply in very restricted ways. Lexical phonology holds that there are not two kinds of phonological rules, such as the general and minor rules of "MPVM" (p. 384), but that there are two kinds of rule application. Applications of rules are distinguished by the domain in which they operate.(6) A rule applies either inside the lexicon, or after the lexicon, or (in some cases) in both domains. Here are some examples of each of these types of application in English:

(1) Lexical application only: addition of -y to stems

such as president, current, or dormant causes

the t to be pronounced [s] (and written with a c)

as in presidency, currency, dormancy, etc. That

this is a lexical phenomenon and not a phonetic

one is shown by the fact that adding the phonologically

identical but morphologically different

[iy] to other words ending in nt does not produce

the consonant change: mint + y does not become

*mincy, and aunt + ie does not become *auncie.

(2) Post-lexical application only: purely allophonic

rules, such as aspiration of voiceless stops initially

in stressed syllables, or nasalization of

vowels before nasal consonants; and assimilation

rules such as the realization of the plural morpheme

as [s] after voiceless consonants and [z]

after voiced sounds. Since these are exceptionless,

predictable, and phonetically motivated,

they belong not in the lexicon but in the postlexical

part of the grammar.

(3) Application both in the lexicon and post-lexically:

palatalization of s before y. Race + -ial and face

+ ial are pronounced [reysel] and [feyzel], and

miss + you is (often) pronounced [misye]. Note

that when this rule is applied lexically, it is exceptionless

(no non-pedant says *[reysiyel] or

*[feysiyel]), but its post-lexical application is optional

or stylistically controlled. It is not unusual

to hear [misye], and if there is a pause between

the syllables, palatalization of [s] is definitely

blocked.

3. Application to Persian verb stems. A venerable rule of consonant assimilation and dissirnilation in Persian states that in a cluster of two obstruents, the first must agree in voicing with the second, moreover, if the second is a stop, the first must be or become a fricative:

[GRAPHICAL DATA OMITTED]

This rule, which produces waxt in relaxed speech f m waqt `time' appears to be a post-lexical rule, although it is by no means allophonic. The past tense morpheme is [t] after obstruents (stops and fricatives) and [d] after sonorants (vowels and all other consonants). Any stem-final consonant appearing before this [t] will, by application of the above rule, appear as a voiceless fricative, as in kuftan/kub `grind' and raftan/raw `go'.(7) One of the characteristics of post-lexical rules is that they do not have information about word-internal information, such as the presence or absence of morpheme boundaries. It might therefore seem that, since it applies both word-internally and across the morpheme boundary between the stem and the past tense marker, the rule should be applicable both lexically and post-lexically. It appears, however, despite its having been in the language for millennia,(8) to be a phonological rule in Persian, one which simply applies whenever its structural description is met, without regard to the presence or absence of morpheme boundaries or past tense markers.(9)

The same consonant assimilation rule applies, but with various changes in stems ending in non-labial consonants, all of which have shifts in their place of articulation. Examples are eighteen verbs like amextan/amez, and a number of verbs in which the past stem ends in s but the present stem in h.(10) The voicing assimilation and fricativization is still attributable to the post-lexical assimilation rule, but the changes in point and manner of articulation are definitely lexical rules, the kind children leam some time after they have mastered the regularly applying phonological rules.(11) These stem alternations are historical relics and should be described with mechanisms which do not masquerade as regular phonological rules.

The large class of denominal and deadjectival verbs,(12) formed by the addition of i to the present stem, is a productive class. Given a new borrowing from English such as modim `modem', Persian speakers could construct a verb modimidan/modim `to (communicate via) modem'. The extra syllable is added automatically, although it is not a phonological process. That it is automatic and productive is shown by such doublets as kustan ~ kusidan `kill'.

We thus have only two classes of verbs, down from five: (1) those whose stem alternations are entirely predictable (invariants like kandan `dig', augmentatives like raqsidan `dance', and consonant-assimilation verbs like yaftan/yab `find'), and (2) those whose alternations are so idiosyncratic as to need to be learned one at a time. This last class includes the truly suppletive verbs such as didan/bin `see' and the partially suppletive class, in which "phonological-look" rules can describe the alternations, but the form of the verbs themselves gives no clue as to which rule is to be applied. Phonological rules apply to forms which meet their structural description. Rules which behave otherwise are descriptive of phenomena which are not phonological.

The advantages of the lexical phonology approach include (1) cutting the number of verb classes of modem Persian from five to two; (2) avoiding the necessity of marking even invariant verbs as "exceptions" to certain rules (cf. "MPVM," 385); (3) clearly separating, on a principled basis, those phenomena for which a child could plausibly leam a rule from those which must be memorized singly, and (4) allowing the use of rules to describe historical relics - once, but no longer, rule-governed - and other oddities, without the implication that those rules have the same status as those describing predictable phonological phenomena.

(1) Michael M. T. Henderson, Modern Persian Verb Morphology," JAOS 98 (1978): 375-88, hereinafter referred to as "MPVN." (2) The count for each class of verbs is from Haim's Shorter Persian-English Dictionary (Tehran: Y. Beroukhim and Sons, (1963). (3) In these examples I shall list the past stem with the infinitival suffix -an, separated by a slash from the present stem. (4) "MPVM" did, I still believe, give a very plausible account of the altemations in the morphemes other than the stem of the Persian verb, even all these years later, I see no need for revision of the rules which "spell" the grammatical formatives. (5) Many linguists prefer to generate grammatical formatives, such as the allomorphs of the English plural or the Persian optative, directly, positing no underlying form. As I showed in "MPVM," I subscribe to this view, which is expressed well by James W. Harris in his Spanish Phonology (Cambridge, Mass.: M.I.T. Press, 1969), ch.3. (6) Antonia Y. Folarin, "Lexical Phonology of Yoruba Nouns and Verbs" (PH.D. diss., Univ. of Kansas 1987), 15ff. (7) The w of Kabul Persian, the variety with which I work, behaves like v in cvery way but its actual articulation. It is pronounced [v] in Iranian Persian. (8) H. Hubschmann, Persische Studien (Strassburg: Trubner, 1895), 114-15. (9) There are some rules which apply only across morpheme boundaries. The coalescence of a and i to long e in Kabul Persian is attested only when the i is the Ezafat, so xana+i+kalan `big house' is pronounced xane kalan. But morpheme-internal instances of ai are realized as e only variably. See Michael M. T. Henderson, "Diglossia in Kabul Persian Phonology," JAOS 95 (1975):651-54. (10) All of these are listed in the appendix to "MPVM." (11) Data from children acquiring Persian natively are lacking and, sadly, likely to remain so for some time. It is conceivable that children may acquire a rule for some of the most frequent consonant altemations, such as past s alternating with present r(d). Experiments in which past tense forms were elicited from children who were given made-up present tense verbs like *meliram could show whether they were likely to produce an invariant such as *lirdam or a rule-obeying *listam. (12) Not all are derivational. There seems no reason to derive xandidan `laugh' from an underlying noun xand - quite the opposite. The formation, distribution, and semantics of Persian verbs have been interestingly discussed by Hassan Zand in "Aspects of Persian Intransitive Verbs"(CPH.D. diss., Univ. of Kansas, 1991).