On phonological representations, rules, and opacity

On phonological representations, rules, and opacity

Lingua 37 (9.975), 363-376 olland Publishing Company Rccelvcd Novt!mhcr I974 A t‘undaIllCntill wnccpt ot‘ standard generative phonology stating that...

2MB Sizes 0 Downloads 91 Views

Lingua 37 (9.975), 363-376 olland Publishing Company

Rccelvcd Novt!mhcr I974

A t‘undaIllCntill wnccpt ot‘ standard generative phonology stating that related morphemes ‘nave uiiique phorK&jpiCai represei;tatiGfis is Criticized . !t is argw!! h! mud! more morphologization of so-called phonological rules is needed to explain certain morphophonemic changes and that rules should be restricted to apply where the morphemic alternants stand in an inflectional (paradigmatic) relationship. Three rules in Israeli Hebrew are discussed, and evidence is brought in from both adults’ and children’s dialects to support the view outlined above. The concept of opacity is examined and arguments against reordering as a means of obtaining grammatical transparency arc offered. The empirical basis underlying evaluation criteria of grammars is emphasized.

ntroduction. Some of the basic claims of standard gener;ttive phonology (SCP) will be challenged in the hght of substantive evidtance taken principally from three sources: (a) frcjm adults’ speech, arguing that they i* c not necessarily aware of the types of regularities that would be claimed by SGP; (b) the question of productivity; and (c) the acquisition of inorphology and phonology by children. By SGP, we basically refer to the assumption, as Skousen (1973) states it, that “regularities in data are to be captured as rules”. He continues: “generative phonologists have, in general, captured regularities that are pased on (I) unique underlying representations and (2) phonetically plausible statements relating underlying representations to surface representations” (p. 72). We shall deal primarily with point (1) and refer to it as the principle of uniqueness. The principle of uniqueness requires that ah related forms of a morpheme have one single underlying representation. Thus, Schane ( 1973: 74) writes:

364

M. Barkui 1 On phonological representations and opacity

‘9. The forma pes (pass) and pack (pack) constitute two separate morphemes which differ phonologically in the final consonarlt. The forms elektrik and elektris [in the related forms electric-electricity respectively: MB] differ in the same way in their final consonants, yet they are variants of a single morpheme. We would like to show that at some level elektrik and ei&ris are somehow the same in a way in which pm and pack are not. If we have an und4ying representation in which both alternants are represented identically, then we have given a unique representation to a unique morpheme.”

The line of investigation to be followed in this article will be, firstly, to examine two rules in the phonology of Israeli Hebrew (IH), placing emphasis on the principle of uniqueness. In a sense, then, this analysis will be given in the guise of the devil’s (devils’?) advocate (sections 3--6). In the second part (sections 7-9), we shall present arguments which cast considerable doubt on the validity of this basic assumption of SGP and all that this entails as a result. In section 6, we shall also raise some theoreticai probiems related to tie data, and shall introduce a third rule.

2, An outline of the verbal system of IH. Before examining the two rules in question, a brief outline of the verbal system is necessary. This system may be divided into seven verb classes (henceforth VCs - in the terminology of Hebrew grammar: binyanim). Each VC has three forms indicating tense: past, present, and future. These tense forms are given in section 3. There are also an imperative and an infinitive which formally resemble the future tense, as well as a nominalized form. The assignment of a verb to a certain VC is arbitrary; in other words, the verb is morphologically conditioned. In general, however, the basic ‘VCsare VC 1 (Qal) and VC 2 (Pi’el)., while the other VCs are derived from these two basic classes. Thus VC 3 is the passive form of VC 2 and ‘VC 7 frequently serves as the passive of WC 1; VC 5 often acts as a caus;;ttive or inchoative aspect of the underlying VC 1 form, and so on. In d:>therwords, the verb may have forms in more than one VC, as will be noted from some of the examples given in the following section. The fist two rules to be discussed are spirantization and n-deletion. As far as spirantization is concerned, we shall concentrate here only on one particular environment: after a prefix vowel followed by a morpheme boundary (henceforth MB% ‘Morpheme Boundary Spirantization’).

M &rtcJi / On phorrol@caI

represert ta tims arrd opacity

365

There are six stops in II-i: /p t k b d g/. Of these, /p/, /b/, and ‘some’ of the /k/‘s undergo MBS; /t/, /d/, /g/, and the ‘other’ /kj’s do not. EJ(;3!-: from the seven VCs are presented below. The forms given I,Aes IL:past, r;resent, and future in the 3rd person masculine singular. The other fcrms do not constitute additionai evidence one way or the other. For those familiar with the terminology of Hebrew grammar, the Hebrew name is given in brackets after the VC, which is assigned a number. We shall make use of this number throughout the article. As stress plays no role in this discussion, it will not be indicated in the data. + indicates a rnor?herne boundary.

VC 1 (Quai)

1. 9 “.

3.

Past

f+esetrt

Future

Gloss

patar katav patax

poter kotev poteax

ji+ftor ji+xtov ji+ftax

‘exempt’ ‘write’ ‘open’

mesfater me+xabes me+vakeb

je+fa ter je+xabes je+vakeE

‘dismiss’ “iaunder’ ‘request, seek’

me-tfu tar me+xubas me+vukag

je+fu tar je+xubas je+vilka!!

‘be dismissed’ ‘be laundered ‘be sought’

mittpater mit+bakeg mit+katev

jit+pater jlt+bakeS jit+katev

‘resign’ ‘was requested’ ‘correspond’ (note 2)

ma+fsik ma+xtiv ma-Wit

ja+tsik

j3+xtiv ja+vlit

‘stop’ ‘dictate (a letter etc.) (2,12) ‘make prominent’

mu+fsak mu+xtav mu+vlat

ju+fsak ju+xtav ju+vlat

‘be stopped’ ‘be dictated’ ‘be protruded’

vc 2 (Pi’@ 4. 5. 6.

pitcr kibes bikef VC 3 (Pu’al)

7. 8. 9.

putar kubaf a)uka!! VC 4 (Hitpa’el)

10. 11. 12.

h it+pater hl t+bakeS Gt+katev VC 5 (HifV)

13. 14. 15.

hitfsik hi+x tiv hi-Nit VC 6 (Huf’al)

16. 17. 18.

hu+fsak hu+x tav hu+vlat

hi. Barkai / On phorlological reprcsen ta tions and opacity

366

vc 7 (;wbl) 19.

ni+ftar

20. 21.

ni’ ?;tilU

ni+fsak

ni+ftar ni+xtav ni+fsak

ji+pater ji+ka tev ji+pasek

‘die’ (1,4,7, 10) ‘be written’ (2, 12:, 14, 17) ‘cease’ (13, 16)

We note that the fricative (f, v, x) follows a vowel followed by a morph< emo,boundary. The vowel is necessary in stating this rule because of the VC 4 forms where there is a stop preceding the morpheme boljndary and there is no spirantization. The morpheme boundary is necessary since spirantization no longer acts in a simple post-vocalic environment (note in the data, 5, 8). The only exception is the future form of VC 7. The MBS may be informally pres::nted as: MBS. p, b, k + f, v, x/v+ -_

4. N-dekion The second rule we shall discuss involves the deletion of the dental nasal consonant in the environment V+_ C. With only one or two lexical exceptions, the nasal is not deleted ir; any other environment. The environment thus involves /n/ when it is the first radical (R, : i.e. the first consonant in the root), whe-AI it 1immediately follows the prefix vowel and precedes the second consonant of -i!le root. Since this environment is never met in VC 2,3, or 4, we can omit these VCs from consideration. We shall also omit reference to VC 7. The forms that interest us are, therefore, VC 1, 5, ,and 6. Since VC 6 represents in capsule form the problems to be discussed in relation to VC 5, we can simplify further by referring only to VC 1 and VC 5. Some examples: vc

1

Past

PLesent

Ftr trrre

Gkoss

22. 23.

. nrrfiil naga

25. 26.

nasa natan

jl+pol jitsa ji+pu jitsa jit ten

‘fall’

2.

nofel nosca nogea nose noten

na sa

‘go’

‘touch’ ‘marry’ ‘give’

There are, however, many verbs in VC 1 where /n/ is not deleted, e.g.:

367

27.

28. 29.

nlrkat ruv3x

ji+nkot ji+nbas (or jinv3.x 1 ji+i&trn

n&am e: I‘

These are rnar~y more verbs with no deletion of /n/ - the reader should not be deserved by the shortness of the list. For VC S the following examples will suffice: 30.

31. 32 33 32

hipil hihi IiIbit hlpia 11lsi3

m3pil IlUh

ii1

illihblt

nxrgia mdsia

jtipil

‘drop, cause to t‘r~ll’(note 22)

jabh

‘t‘xprt‘ss,

jrhit jupia jasia

‘look rrt’ ‘reach’ (note 24) (give someone a ride’ (note 23)

state’

There is a number of verbs however, where /n/ is preserved (always):

35. 36. 37. 38. 39.

Yast

h?scrr t

h

hinbit hinpik

manbit manpik rnanmix manciax manSim

janbit janpik janmix janciax jar&m

hinmix

trinciax llinSim

lure

Gloss ‘blossom, bud’ ‘allot, issue’ ‘lower’ ‘perpetuate’ ‘apply artificial respiration’

Two -questions may be raised with regard to the /ni as R, in the above data: (i) Which morphemes should be considered exceptional -- those deleting /n/ or those preserving it? Having decided upon an answer to the first question, we may ask (ii) What mecha SPIbest accounts for the exceptionality? As regards tl x first question, we note that the deletion of ini is r,ot productive. New verbs coming into the language maintain /n/ - indeed, most of the verbs as illustrated in 35-39 above have only recently been introduced into the language. Furthermc,. e, verbs which historically had deletion, have in some &es had /n/ restored to them, e.g. Biblical Hebrew (BH) jiyqom (with total assimilation), IH: jinkom ‘he will avenge’; BH: jicldlyr, IH: ji&<>Y‘he will vow’, etc. It will be noted from the BH examples that there was total assimilation of /nf to the following consonant. In IH, however, there is no reason whatsoever for referring to a rule which totaPly assimilates /n/ and we shall regard it as a case of simple deletion.

368

M. Barkai / On phonological representations and opacity

In the light of the above facts, it seems clear that the deletion rather than the preservation of /n/ is the marked case. In answer to the second question raised, it is proposed that the best approach for capturing the exceptionality of the n-deletion is to consider this rule a minor rule (Lightner 1968; Lakoff 1970) and to mark those morphemes which undergo the rule with a specific feature: [+ndeletion]. Morphemes not undergoing the rule carry no specific marker.

5. The ordering relationship between the two rules

It is clear that the ordering relation between those two rules is (1) MES, (2) n-deletion. Thus, to /ji+npol/ (ex. 22), MBS would not apply and then n-deletion would give [jipol]. From /hi+nbit/ (ex. 32) the structural description of MBS wild not be met if applied before n-deletion, which then leaves us *withthe correct phonetic output [ hibit]. To sum up, we have referred to two rules. A MBS rule which converts jp/, /b/ and /k/ into their corresponding fricatives after a vowel followed by a morpheme boundary. The other rule is n-deletion, which applies in a similar environmem, i.e. after a vowel followed by a morpheme boundary, provided it also immediately precedes a consonant. This rule is, however, limited to certain morphemes.

6. N-deletion in VC 5 forms.

We have been assuming throughout that the stems of verbs like [hipil], [hibit], [hisia] all have /n/ as R, (i.e. as the first radical) immediately preceding R,, i.e. /hi+nCiC/. In the case of [ hipil] (likewise [hisia] and other verbs) we cited the clear relationship - both semantic and phonological - this form has with VC 1 forms like [nafal]. Henceforth we shall refer to cases ,such as these as ‘related morphemes’. In these cases, we may say that the correct phonological representation is /hi+npil/. A much more compler; situation arises with verbs like [hibit] which have 110related morphemes (henceforth ‘isolated morphemes’) surfacing with an /n/. What justification is there in setting up an underlying form with In/, i.e. /hi+nbit /? Kiparsky (19681 argued that neutralization processes cannot apply to all occurrences of a morpheme (this is the stronger version of the ‘Alter-

M. &rkai / On phonological represen tntions and opacity

369

nation Condition’). According to this condition, which is a constraint on the Ilature of underlying representations, we would reject the underlying rep,esentation /hi+nbit/ since /n/ is neutralized everywhere as zero in this morpheme. We therefore should posit /hi+bit/. Such a representation WO!I 4d apparently require us to state that this verb is an exception to the MI33jr-ule. In terms of lexical features, we have traded in one exception feature for another, for n-deletion, as has already been shown, is a minor rule. So, in terms of cost, /hi+bit/( MBS is just as marked as /hi+nbit/l+,,,l 1 Hcawevcr, this analysis is unsatisfymg. 71lis is because there are now two ’ different reasons as to why spirantization does not apply: in isolated morphemes, because they are exceptions to the rule; and in related morph12mes, where the ‘failure’ of spirantization can be accounted for by orderirrg NIBS he OTC> n-deletion. But surely the reason for the s-tirface bi-labial stops is the same for both types of verbs. Let us consider some further data which are relevant to the problem being discussed:

40. 41. 42: 43 44. 45.

Past

iken

t

he+fic he+vin he+fii he+vis he+rim he+&

me+fic me+vin me+fik me+vis me+rirn me+sir

Future

moss

ja+fic j a+vin ja+fii ja+vis ja+rim ja+sir

‘distribute’ ‘under:: tand’ ‘distribute’ ‘defeat ’ ‘raise’ ‘remove’

The analysis which recognizes an underlying /n/ in verbs like [ hibit 1, [hipil], etc. helps us to account neatly fo he verbs presented in 40-45 for two reasons: in these verbs, we note ( that the bi-labial stop is always spirantized, and (ii) that the past tense vowel i is realized as e, as is the present prefix vowel a (henceforth we shall refer to this process as ‘midding’). In the latter case we can say that ‘midding’ applies before a single consonant before a vowel. Thus, since [hibit], etc., has been analysed as /hi+nbit/, there being two consonants following the prefix vowel ’ A rather nasty problem arises, however, from the fact that we ould posit RI as the dental lateral /I/ in verbs like [hibit]. /Ii is also deleted in exactly the samt: environment ir; the VC 1 verb [lakax) ‘he took’ but future [jikax] from /ji+lkax/. Thus, mt also need a minor rule for this lexical item. Since n-deretion is also a minor rule we lose nothing in terms of cost by positing /hi+lbit/ and marking it as [ +ldeletion 1. That this would surely be rejected by a Hebrew phonologist is.not to be doubted - but would this rejection be on synchronic grounds? To put it another Way: can /hi+nbit/ really be defended on synchronic grounds?

’ y

370

M. Barkai / On phonological representations and opacity

we need only order the rnidding rule before n-deletion to generate the correct output. Up to this point, we may summarize our (tentative) analysis as follows by showing the following derivations: Phonological representation: Midd ing : MBS: n-deletion: Phonetic output:

/hi+nbit/ [+n-cieletion] hibit [hibit j

/hi+npil/ (+ndeletion) hipil [ hipil]

/hi+pic/ hepic hefic ;;fic,

As far as the ‘midding’ process is concerned, we have been considering what may be regarded as a rather conservative dialect of IH. There are, however, other dialects which should be considered. The simplest cases, from our point of view, are those dialects (henceforth Dialect ‘A’) which do not have the middlng rule at all. The verbs listed in 40-45, in this case, have the same prefix vowel as [hibit-mabit], i.e. [i] in the past and [a] in the present. We c )uld say that, as the midding rule is contradicted on the surface because of the verbs derived from R, = /r-r/, it is subject to rule loss (Kiparsky 147 1: 627-8). It should be noted, however, that in this dialect there is now only one reason for positing R, = /n/ in verbs like [hibit] (but not [hipil]) - namely to block MBS. Dialect ‘B’, however, has the verbs analyzed as R, = /n/ surfacing wit3 the prefix vowels [he] in the past and [me] in the present, i.e. the vowels which are prese:lt in this dialect in the bi-consonantal verb. To account for this dialect we could argue that n-deletion is reordered to apply before the ‘midding’ rule. In the conservative dialect, the order (1) n-deletion, (2) midding is the marked counter-feeding order; the innovating dialect reorders to the unmarked feeding order. Note, too, that in this dialect we again have only one reason for positing /n/ in verbs like [hibit] - that is, to block MBS. It should further be added that both Dialect ‘A’ and ‘B’ are probably somewhat mythical. It is not likely that there are speakers who consistent ly regularize the VC prefix one way or the other. Nevertheless, these two dialects are discussed because the possibility of their existence cannot totally be ruled out. There is yet another dialect (Dialect ‘C’) which is n >t nearly so ‘neat’ as the three dialects discussed above. In this dialect, some R, = /n/ verbs may be like bi-consonantal verbs, and some of the latter may be like the former. Even more vexing is that some verbs of either group may well

be in free variation, e.g. [hipil] - [ hepil] , [ hific] - [hefic], etc. Within the tItleoretica1 framework we have been discussing we would have to mm to the conclusion that some Hebrew speak~s (they almost certainly constitute the majority) reorder one way otl one occasion and in llnother way on a different occasion and/or that the midding rule is an optional rule; and, furthermore, that some verbs will have to be marked YSPDt undergoing the midding rule at all. To illustrate assume a speaker ‘& the f4lowing surface forms: (i; (hipil] - [hepil], i.e. in free variation; in this case we may assume either (a) that the speaker sometimes chooses one order ((l), midding, (2) n-deletion to generate #pill) and sometimes the reverse order (for Jill): or preferltbly (b) that n-deletion is ordered before midding, which I optional rule. (ii) fhific]: this form would have to be marked as not undergoing the optional rule of midding. This verb would be appropriately -marked in the lexicon. (iii) [hivis] - [hevis] (a bi-consonantal verb): midding can be marked as an optional rule in the grammar (as argued in (ibJj above.) (iv) [hevin]: midding applies obligatorily, and an appropriate mark is assigned to this verb in the lexicon. “Whatthis dialect indicates is that the midding rule is optional and furthermore that it is a lexically marked rule. The role of initial radical /n/ is from the phonological point of view, reduced to being a ‘protector’ of the bi-labial stops. One further question must be raised: we have noted that in the innov;rGng Dialect C, a speaker can have an output of, say, [hepil] or [ hebit] (with or without free variation in the prefix vowel). In this dialect we might wish to claim that n-deletion has been re-ordered so as to feed the (optional) rule of miciding. But in r~ ialect is there any tendency whatsoever to re-order n-deletion and IL! ven though they stand in exactly the same counter-feeding order as do n-deletion and midding. (If such re-ordering had taken place we mig!lt have expected forms like * Izifi’l (* hc’fi’l); * hivir (* hit), etc.; such forms are totally unacceptable). Before offering an answer to this problem, we should recall Kiparsky’s definition of opacity and what he predicts should be a resultant cl=ngt= in the grammar. Since we are only interested in the first of the two conditions he cites, only this will be given here. Kiparsky writes (197 1: 62 l-2) that “a rule A-+B/C D is opaque to the extent that there surface representations of the firm . .. A in the environment C - D”. The

372

M. Barkai / On phonological representations and opacity

opacity may be gotten rid of, i.e. the grammar may become transparent (in Kiparsky’s terminology) either (i) by the rules being reordered; or (ii) by the opaque rule being lost from the grammar. Bearing the above in mind let us return once again to the conservative dialect whose rule ordering reflects history. It should be noted that there is no ordering relation between MBS and midding. Proposed underlying representation: /hi+npil/

/hi+pic/

[ +ndel. ]

MBS: p, b, k-+f, v, x/V+Midding: i, a-+e/_ + CV N-deletion: n-@/V+_C Phonetic output:

hipil [ hipil]

hific hefic x[hefic]

In terms of opacity, we n Jte the following: (i) Both MBS and midding are contradicted on the surface and they are both opaque. The opaLity in both cases is caused by the n-deletion rule. The ordering relationship is the marked counter-feeding one. The expected change would be a re-drdering to the unmarked feeding relation, so that n-deletion will precede MBS and midding. What we find, as previously stated, is evidence for n-deletion preceding midding (giving us innovating forms like he@, kebia, etc.), but no evidence whatsoever for n-deletion being re-ordered before NBS. (ii) N-deletion is contradicted on the surface not by virtue of its rzlationship to any other rule but because of the exceptions to the rule itself (e.g. verbs 27-29 (VC 1) and 35-39 (VC 5). We might perhaps expect loss of the rule from the grammar or perhaps its restriction on morphological grounds. The main problem (relating to point (i) above) is, when there is the same marked counter-feeding relation caused by the same rule (n-deletion), why then should Kiparsky’s predictions be borne out in one case (n-deletion and midding) but not in the other (n-deletion and MBS)? To aiswer this question we could say that: (a) Kiparsky’s conception of opacity needs modifying; and/or (b) the grammatical theory, and the types of underlying representations and ruin it requires, is at fault. It follows that if(b) is the case, (a) will probably be true also.

h?

Batkai

/ On

phonological

representations

nrzd opacity

373

7. A revised andivsis I

Wow then t:an we explain the data? The changes which are taking place ebrew today are best explained b> claiming that underlying reprere not so unique or abstract as has been claimed by SGP. If we impose paradigmatic constraints based on inflectional relationships 0 fr ~~Arderlyir,g representations then we can explain the phenomena in question. In Hebrew, the paradigmatic constraint appears to be based on tire VC itself and the inflectional relation is basically one of tense i.e. among the three tenses which constitute the VC, namely, present, past, ;Ilndtkture. In other words, what we must examine are the alternations whicl~ exist within the VC itself. Let us re-examine the data in the light of t hcse proposed restrictions.

In VC 1 there is an alternation such that the stop occurs in the past and present tense forms, the corresponding fricative in the future (see cx. l-3). Therefore, both may be derived from one common source. In VC 5, however, there is no alternation whatsoever. Either there is always a stop (examples 30-32) or there is always a fricative (e.g. 13-15 O-43). Therefore MBS no longer applies in this VC. In other words, the rule must be morphologically conditioned so as to apply only to those VCs in which there is a stop: fricative alternation. It follows that since MBS does not apply in VC 5 that there can be no re-ordering between this rule and n-deletion - hence we can explain why forms such as * hefil * (hifil), * J:e& “(hivirl are totally unacceptable. 7.2. N-deletion he position regarding this rule parallels what we have sr,id about MBS. IVC 1, there is an alternation (ex. 22-26) therefore n-deletion applies. Because of the large amount of counter-examples (a few are cited in D---29), this rule is a minor rule. In VC 5, however, there is no n + 8 alternation. Either (view!:d synchronically) there is no ii/ at all, e.g. 30-34, or In/ is always maintained (35-39). It follows tha we cannot explain the changes relating to midding by claiming a reordering with a rule which does not apply. This view also explains why the verbs with /n/ (35--39), which are all

374

M. Barkai / On phonological representations and opacity

new creations, present no learning problems. They can be -,reated as the n-deletion rule does not apply to this VC. 7.3. Midding This rule is relevant only to VC 5. Within this class, we find surface representations sometimes with /i/ (30-34) and sometimes with /e/ (40-45j before the environment CV. To know when [i] occurs and when fe], must simply to be learned. In other words, we argue that this rule is governed by idiosyncratic markers in the lexicon. This explanation renders the rule exceptional in a similar way to n-deletion in VC 1: hence the opacity. Once again it follows that there can be no question of re-ordering as an explanation for the i-e variations, since there is no relevant rule with which it can be ordered in the first place.

8. On Children

It is interesting to pause for a moment and see how children handle the three rules in question. As we might expect, children have the same performance as adults do ‘and more’. The ‘more’ refers to the difficulty in mastering the n-deletion rule in VC 1. This is what we would expect since the rule is opaque. Children have a tendency of suppressing the rule altogether. Thus, the usr=forms such as [jinsa] ‘he will travel’ for [jisa], inpol] ‘he will fall’ for [jipol], [jinga] ‘he will touch’ for [jiga]. They appear to be able to master the rule and delete /n/ by approximately the age of 9 or 10, although [jinsa] has been heard by me on occasion even in adults’ speech though this would be regarded as sub-standard. The ‘midding’ rule, as we might of course expect, causes great difficulties for children (see Bar-Adon 1959: 57-9). Three reasons may be off’ered for why children master n-deletion before midding (if they ever come to master this latter rule!): (i) the number of lexical items to which n-deletion ;jpplies is very small and the verbs involved are cornmanly used; tii) the n-0 relationship is c1early easier to perceive than the a/i-e relationship at the phonetic ‘fevel;and (iii) the allomorphs /ma-me/ or /hi-he/ have different realizations in the same environment, whereas the n-8 alternation is realized in quite different environments. The purpose of mentioning this point here is to suggest what factors may be involved in the child’s jmastery of opaque rules, fc:,, as Kiparsky ( 197 1: 622) rightly

M Burkai / On phonokugicalrepesen tations and opacity

375

says, ‘opacity .. . is a matter

of degree’. They may be summed up as (i) frequency of application of rule and frequency of use of the lexical items involved; (ii) the phonetic basis underlying the alternation; (iii) environmental factors.

It is clenr from our arguments that the notion of unique underlying rel jresetltations has to be seriously reconsidered. We have shown that it is llot af all plausible in the light of the ongoing changes and the way in which

children build up their grapinlars to claim that, say, /hi+npil/ must 11~‘the tinderlying form in VC 5 i.e. with an /n/ present because it is reElatedto forms where In/ surfaces, such as in [natal]. -Wehave argued that the underlying representation for [3kil] (or [hepil]) is /hi+pil/ which

means that we have increased the amount of information required in the lexicon, since we have both rzyl and pl as lexical roots 2. They clearly are associated in some way, our argument being that the association is nat a pl~OnolOgiCi~llysignificant one. Notice too that we have concentrated only on the consonants, i.e. we have argued that SGP requires /hi+npil/ and the like because of the principle of uniqueness - but what of the vowels? What is their unique under lying representation? Note that if we go back still further to the lexical representation of a purely consonantal root and claim that here lies the uniqueness then we are faced with another very serious theoretical problem, since purely consonantal roots violate naturalness conditions (see Postal 1968: 53--97) on underlying representations. Further research may show that these naturalness conditio s are basically anti-Semitic, and therefore have no place in linguistic t eory. Pandora’s box thus operled must be left for further research. It should, of course, be pointed out that we cannot invoke some a prioi-i evaluation measure to decide whether a grammar with more concre&c phonological representat ions ? and fewer rules (though with more morphological conditioning) is more complex or simpler than one with 2 Schwarzwald (1973)independentlyarriveda; a similarconclusion. 3 Wehave been tacitly assumingthat if thereis a root then it is registeredat the level of lexical representation. The ‘concrete’ phonological representationsare those discussedthroughout the paper and would presumablybe the output of morphologically conditioned vowel intercalation rules. See Schwarzwald 1973 for furtherdiscussion.

376

M. Barkai / On phonological representations and opacity

unique representation and more (phonologically conditioned) rules. Ultimately, the evaluation measure is to be decided upon on empirical grounds, (see Chomsky and Halle 1968: 330-2). In this article, we have presented empirical evidence based on children’s grammars and on a morphophonemic dhange in progress (midding) supporting ‘concrete’ non-unique phonological representations, more lexical marking, and fewer rules, as opposed to unique representations, fewer lexical markings, and more rules. To put this in psychological terms, we holci that the evidence supports the claim of greater memorization and of simpler and more direct retrieval by the speaker-hearer of a language as opposed to the converse view inherent in the ‘abstract’ (SGP) approach. Braine has recently argued that “a totally different kind of research strategy, one directly addressing the psychological reality of representations and rules, is needed to determine the kinds of constraints operative on lexical representations - in particular, the degree of abstractness of phonological units” (1974: 298). The discussion presented in this article is a step in that direction

References Bar-Adon, A., 1959. Leshonam hameduberet she1 hajeladim bejisrael [The spoken language of children in Israel]. 2 vols., Jerusalem: Hebrew Univ. (doct. diss.). Braine, M.D.!%, 1974. On what might constitute learnable phonology. Lg. 50(2), 270-99. Chomsky, N. and M. Halle, 1968. Sound pattern of English. New York: Harper and Row. Kiparsky, P., 1968. How abstract is phonology? Unpubl, MS. Kiparsky, P., 1971. Historical linguistics. In: W.O. Dingwall (ed.), A survey of linguistic science, College Park, Md.: Univ. of Maryland I inguistics Program. Kiparsky, P., 1972. Explanation in phonology. In: S. Peters (ed.), Goals of linguistic theory, Englewood Cliffs, N.J.: Prentice-Hall. Lakoff,G., 1970. Irregularity in syntax. r;‘d:wYork: Holt, Rinehart and Winston. Lightner, T.M., 1968. On the use of minor yules in Russian phonology. J, Linguist. 4,69-X. Postal, P.M., 1968. Aspects of phonological theory. New York: Harper and Row. Schane, S.A., 1973. Generative phonology. Englewood Cliffs, N.J.: Prentice-Hall. Schwarzwald, O., 1973. Lexical representations, phonological processes and morphological patterns in Hebrew. Houston: Univ. of Texas (doct. diss.), Skousen, R., 1973. Evidcllce in phonology. In: C.W. Kisseberth, (ed.), Studies in generative phonology. Champaign, Ill. : Linguistic Research Inc,