Neurological Evidence for Lexeme-Morpheme Base Morphology

Robert Beard (1986)

Acta Linguistica Academiae Scientiarum Hungaricae, Tomus 36 (1-4), pp. 3-23 (1986)

Since its inception, the generative school of linguistic theory has struggled to establish an attitude toward the relation of its competence model of grammar to a psychological model of performance (parser). Initially it was hoped that the relation would be "transparent", that the items and operations of grammar would be isomorphic with those of the parser. However, early attempts at confirming the psychological reality of transformations came to naught and Chomsky retracted his position to one which allowed the opaque, homomorphic relationship developed by Fodor, Bever and Garrett 1974.

More recently Bresnan (1978) and Berwick and Weinberg (1983, 1984) have developed an argument for using the performance model as a tool for the evaluation of the adequacy of theories of grammar, claiming that the more transparent the relation of grammar to parser, the more adequate the model. Berwick and Weinberg call this the "Type Transparency Hypothesis". Bresnan feels that in order for type transparency to work for grammar we must abandon transformation rules in the absence of evidence that they operate directly in speech and in view of the linear operation of parsers. However, Berwick and Weinberg demonstrate that Bresnan's Lexical Function Grammar, which accounts for transformational relations in the lexicon, is superior by this test only assuming a cognitive mechanism capable of carrying out a single operation at a time. If we assume a more reasonable model of human mental capacities, a mechanism which can carry out several operations simultaneously, the various levels of ST, EST, REST and even GB can all be accommodated in a transparent parser. They propose such a model, following lines originally laid out by Marcus (1980).

In this paper I will demonstrate that given these latest assumptions about the nature of mental processing, the morphological model which receives the greatest support from the neurological evidence is one which separates morphology from the lexical and syntactic operations it marks (the Sepa1·ation Hypothesis (SH) ) and distinguishes lexemes and (grammatical) morphemes as discrete natural classes situated in autonomous components of grammar (the Lexeme-Morpheme Base Hypothesis (LMBH)). Section 2 outlines such a model. Section 3 reviews the "classical theory" of aphasia, a syndrome of speech failure associated with distress of the speech regions of the left cerebral cortex. Section 4 then demonstrates type trans-parency between this model and the parser with evidence from research in aphasia. It then goes on to show how one series of recent attacks on the classical theory of aphasia suggests extensions of that theory which in fact strengthen the aphasiological support for SH and LMBH.

Recent interest in linguistic (Chomsky 1981) and cognitive modularity (Fodor 1983) has influenced a growing sympathy for separating morphology from lexical and syntactic derivation in the literature (Anderson 1982, Beard 1976, 1981, Klavans 1985, Pesetsky 1985, Pounder 1986, Sadock 1985, Szymanek 1985). In my version of the Separation Hypothesis, morphemes are purely phonological operations independent of any and all lexical or syntactic operations on grammatical functions of the stem.

The best argument for this hypothesis is the fact that the conditions on lexical and syntactic derivation are independent of those on affixation. Conditions on morphology are related to derivations and the grammatical features they manipulate only in that such features, added previously to a lexeme, form some part of the conditions on morphological marking. The English subjective (agentive) nominalization seems to operate on verbs with few real restrictions. A small, lexically marked group of mostly intransitive verbs, e.g. the copulas be, seem, become and a few others, stay, fall, weigh_[-Tr], *be-er, *seemer, *becom-er, *stay-er, *fall-er (?fall-ee), *weigh_[-Tr]-er seem semantically prohibited from subjectivization. However, neither intransitive

are strictly excluded. Acceptability depends on the availability of a generic referent, i.e. a class of conceivable referents which the derivate might type name, but the output potential seems to depend on no more than the capacity of the underlying verb to support a subject (*rain-er, *snow-er). In any event, marginally attested forms and related compounds like com-er (late-comer) and go-er (church-go-er) demonstrate that the restrictions on such usage are unrelated to the ability to combine with -er.

Constraints on affixation are more elaborate and different from those on derivation. Initially, it might seem that affixation parallels derivation in that all subjective derivations tend to receive -er regardless of transitivity. Intransitive Latinate verb stems, however, consistently receive the affix of the agentive adjective whatever it might be.¹

(4) Latinate (Stative) Intransitive Subject Nominalizations


Derivate	Marker	Derivate	Marker
particip-ant	(in)	(e/im)migr-ant	(from/to)
adher-ent	(to)	descend-ent	(from)
registr-ant	(for)	matricul-ant	(at)
degener-ate	(from/to)	devi-ate	(from)
resid-ent	(at)	operat-ive	(in)

Subjective nominalizations based on intransitive, especially stative verbs, presently tend to receive -ee, the same suffix used to mark animate (direct and indirect) objective nominalizations.² This suggests that affixation alone is determined by agentive and patientive features, without reference to the subjective/objective derivational distinction.

Notice that even if this rule is unproductive, the argument goes through. (6) demonstrates that omission would be the preferred marking regardless of the status of the derivation rule. This marking is contingent on the presence of the particle, not the derivation, for all count nominalizations of verb-particle couples are omissively marked, e.g. these both locative and perfective nominalizations a get-away, a turn-around, a drop-off, and a walk-in, which are also subjectives.

The contemporary morphology for the subjective nominalization requires (i) the corresponding adjective affix for lexically marked intransitive Latinate verbs.⁴ Otherwise, (ii) the subjective nominalization of intransitive (stative) verbs is productively marked by the suffix -ee, which as Carrier-Duncan (1985, 32-3) points out, is an animate "nonagentive" marker. Germanic verb-particle stems (iii) are marked with omissive morphology and (iv) elsewhere the suffix -er marks the subjective nominalization in English. None of these conditions correspond to the conditions on the subjective or, for that matter, any putative agentive nominalization. All the lexical derivates of (1-6) may be generated by a single subjective nominalization rule with virtually no constraints; however, the constraints on affixation are unexpectedly complex, laced with rich subregularity having no bearing on the lexical derivation.

A clear implication of (7) is that (grammatical) morphemes and lexemes are radically different types of entities (Beard 1986). Unlike lexemes, grammatical morphemes and the derivational and inflectional functions they mark fall into separate closed, grammatically ordered classes. While subjective (athlete) and objective (victim) lexical classes are common, they are not related to each other via formal paradigms as are the lexical derivational subjective and objective.

(8)	Subjective	Subjective	Objective	Modalic
	(Agentive)	(Agentive)	(Patientive)	(Instrumental)
	Adjectival	Nominal	Nominal	Nominal

-ee	. . .	escap-ee	draft-ee	. . .
-er	[redder]	dry-er	"keep-er"⁵	mix-er
-ist	reform-ist	record-ist	. . .	. . .
-ant	depend-ent	migr-ant	rehabilit-ant	stimul-ant
-Ø	discriminate	cut-up	slice	lift

(8) does not represent a problem of "polysemy" or "homophony". Lexical polysemy and homophony (or homonymy) are usually defined strictly in terms of the practical problems faced by lexicographers in deciding which dictionary words deserve independent entries. The only objective argument for homophony is historical, e.g. the spelling of pair : pear (homonyms) vs. that of the various meanings of dope, generally considered a polyseme. In fact nothing is gained in defining these terms over a theory of the lexicon which represents the sound-meaning relation as in (9).

The important point of (9) is that whether one sees "homophony", "polysemy" or both here, no difficulty arises in distinguishing individual lexemes as signs in some elaboration of the term. Wierzbicka (1985, 71-73) refers to this definitional characteristic of lexemes as the "discreteness" criterion of the Aristotelean Categorical View of language.

In (10) the four highly productive suffixes of (8) are listed in their complex associations with four highly productive lexical derivational categories. The interesting aspect of (10) is the impossibility of separating these morphemes into discrete signs like lexemes. Therefore, in order to maintain the discreteness criterion, we must conclude that the phonological and functional units involved are themselves discriminate and do not form discrete isomorphic associations. This entails a definition of grammatical morphemes wholly unrelated to that of lexemes.

The argument for discrete lexemes and morphemes rests on three definitional characteristics of lexemes which morphemes do not share (Beard 1986). Perhaps the most important is that lexemes do not exhibit zero phonology. Morphemes may alternate with nothing because all information necessary for semantic interpretation is in the lexeme to which it accrues or in the morpheme's paradigm or is added to the base lexeme by derivation rule. Morphemes also do not occur in phrases independently, derive, subcategorize for case and the like, for morphemes mark these relations among lexemes. Finally, morphemes are formal representations of lexical and syntactic grammar; they therefore belong to the closed classes of grammar and have no meaning at all except in the lexemes whose syntactic and lexical relations they mark. Whatever lexemes are, morphemes are not; whatever omission, reduplication, revoweling, umlaut are, so are morphemes.

(11) A lexeme is a minimally distinguishable (bound or free) ordered sequence of phono1ogica1 segment associated with at 1east one sense, which may be lexically extended by lexical derivation and which belongs to an open, unpara¬digmatic class.

This definition covers the major class items which may fill syntactic nodes: N, V, Adj and Adv. It obviously is not sufficient in the absence of an explanation of homophony and/or polysemy (9). Adpositions are clitic or free morphemes in a class with declensional endings (Beard 1985). An "open" class may be defined here as "synchronically accessible", i.e. a class susceptible to expansion by synchronic operations and borrowing. A "closed" such class would be "synchronically inaccessible" in the same sense. "Lexical extension" here refers to derivational features like [Subject], [Object], [±Feminine], added to the sense, and not to any phonological features added to the formant of the lexeme. (See Beard (1981) for details of "lexemic extension".)

(12) A morpheme is the smallest phonologically distinguishable (bound or unbound) formal modification of a lexeme, which is paradigmatically ordered in a closed class, marks grammatical functions and is not susceptible to L-derivation.

A lexemic "modification" might be a free adjunct attached to a peripheral member of a phrase to mark some grammatical function of the head in the sense that the in the (old) man is a modification of (old) man or it may be an internal or external, bound modification as in man : men or bay : bays. Nothing grammatically substantive clings to the "bound-free" distinction (see also Carstairs (1981) and Marantz (to appear). Whether features like [Causative] and [Potential] are added to verbs by free-standing auxiliaries as in English or appended as affixes as in Turkish is a superficial issue unrelated to the universal issue of why these features and not others are found in languages and how they are manipulated by speakers. The crucial issue is that morphemes are associated only with internal grammatical referents and, usually, only in some context.

The LMBH model of morphology described thus far rests on three major hypotheses about the nature of the morphological component of grammar—all in conflict with some aspect of prevailing sign based theories. First, SH implies that the nature of and conditions on rules which modify a lexeme in any manner to mark grammatical functions are independent of those which add abstract lexical or inflectional (morphosyntactic) features to lexemes. This allows the possibility of discriminate damage to syntactic, lexical and morphological processors, specifically the possibility of the presence of syntactic and/or lexical category features in the absence of morphemes marking them and vice versa, our first prediction.

This prediction is markedly at odds with the predictions of the Saussurian Sign Base Hypothesis, which assumes that all linguistic sound and meaning are isomorphic, biunique, form mutually implied association. Sign-based morphologies, under the Type Transparency Hypothesis, predict that the same processor which computes phonotactic sequences computes the meaning(s) or function(s) associated with them; meaning is accessible only via sound, sound only via meaning. These morphologies imply that a speaker who cannot utter the phonology should not be able to calculate its functions and those who can, should be able to.

Second, if the sound-meaning relation of the morpheme is wholly different from that of the lexeme, we would predict radically different neurological organizations for lexemes and morphemes. The distinction of lexemes and morphemes might come in the form of independent storage at different physical sites on the cerebral cortex, different locations for the access mechanisms of the parser or operationally different access mechanisms.

Discrete storage locations would support the model in (7) but would imply that the Lexical Morphology Hypothesis (Bloomfield 1933, Bresnan 1982, Lieber 1981, Marantz 1984), which situates lexemes and morphemes identically in the lexicon, would have to be modified or surrender any claim of type transparency. Different locations of access mechanisms or operationally distinct mechanisms would not necessarily undermine the Lexical Morphology Hypothesis if accompanied by independent evidence that morphemes and lexemes are stored in the same location. Thus we must keep in mind the distinction between storage areas and retrieval mechanisms, for this distinction underlies the crucial test of the claims of the LMBH, on the one hand, against those of Lexical Morphology, structuralist and semiotic models, on the other.

The third notable aspect of (7), the Integrated Morphology Hypothesis, is supported by the work of Beard, Halle, Lieber, Marantz and others a single morphological component operates on both inflectional and lexical derivational outputs. Although the operations generating abstract derived lexemes and phrases may be discrete, for the Integrated Morphology Hypothesis to hold, the operations modifying lexemes and phrases to mark them for these operations will be conducted by the same component, or processor. Thus we should find at some level, indiscriminately affected lexical and syntactic derivational morphology in aphasia, dyslexia and normal speech error.

The competing Split Morphology Hypothesis (Anderson 1982, Perlmutter 1986), argues that inflectional morphology applies after all morphosyntactic adjustments have been made but that derivational morphology is an independent lexical matter accomplished in the lexicon before category selection and insertion. Again, assuming type transparency, deficits in the two morpho-logies should not occur as symptoms of the same syndrome but should be affected by incidents in distinct cortical areas and should be associated with different cosymptoms if this hypothesis goes through. Moreover, since derivational morphology is assigned to the lexicon, deficits in this morphology should be associated with lexical rather than syntactic deficits.

These are not the only predictions implied by (7) and the full theory accompanying it. The fact that morphemes are operations rather than prespecified items predicts that in the performatively most transparent case morphemes will not be accessed as lexemes are. Since the definition of lexemes in this theory depends crucially on an ordered sequence of phonological segments, left-right (initial-final) ordering errors should predominate in the pathological and other speech error data. The L-rules of this theory are generative and should therefore affect lexical retrieval. How these predictions might emerge in psycholinguistic or neurolinguistic data it will not be pursued here even though the pertinent data are very suggestive.

3. The Classic Theories of Aphasia

The two major linguistic cortical processors seem to be controlled by the anterior or "Broca's" area and the posterior or "Wernicke's" area of the left cerebral cortex. Because damage to it affects the processing of "closed class", "little" or "function" words, affixes and their functions, Broca's area is closely associated with an aphasic syndrome called "agrammatism". The central symptoms of agrammatism may be summarized as

(iv) omission of arguments, e.g. subject and direct object, and misordering of syntactic constituents. Agrammatic patients exhibit little difficulty in accessing lexemes, i.e. (V,) N, Adj, Adv, but omit or find great difficulty in deploying grammatical functors. This type of speech is also called "nonfluent" because of the co-occurrence of dysprosody, the loss of control of intonation, which lends language the semblance of fluency.

Sensory aphasics, on the other hand, who suffer damage exclusively to Wernicke's area, become "fluent" aphasics. Sensory aphasics retain control over closed class functors and intonation, but have difficulty in recalling lexical bases. They speak in a normally intoned stream of grammatical markers, e.g. pronouns, prepositions, articles and auxiliaries, that is, in "contentless" sentences. They may recall lexical items related or unrelated to target lexemes or create nonsense neologisms (paraphasia) one form exhibits a particular difficulty in retrieving nouns (anomia). Thus while it remains a point of contention whether "control" refers to storage, retrieval, processing or some combination of these, Broca's area in some sense controls grammatical morphemes and/or the syntactic structures they mark to a significantly greater extent than Wernicke's area while the latter controls lexical bases to a greater extent than Broca's. However, as Cvetkova and Glozman (1978b) emphasize, traits of agrammatism are found to some extent in both types of aphasia.

C. B. Uh, well this is the ... the /didiŋ/ of this. This and this and this. These things going in there like that. This is /sen/ things here. This one here, these two things here. And the other one here, back in this one, this one /gi?/ look at this one.

C. B. I can't tell you what that is, but I know what it is, but I don't know where it is. But I don't know what's under. I know it's you couldn't say it's .. . . I couldn't say what it is. I couldn't say what that is. This shu—that should be right in here. That's /béali/ bad in there Anyway, this one here, and that, and that's it. This is the getting in here and that's the getting around here, and that, and that's it. This is getting in here and that's the getting around here, this one and one with this one. And this one, and that's it, isn't it ? I don't know what else you'd want.

4. Aphasiology and Current Morphological Hypotheses

The classic theory of aphasia offers obvious support for distinct lexemes and (grammatical) morphemes. With few exceptions (discussed further on), physical loca¬tions in the cortex seem to control discrete classes of linguistic elements corresponding closely to the lexemes and morphemes of LMBH. Moreover, the preponderance of available evidence suggests that these areas are the sites of storage and computation.

Unless we wish to consider all the linguistic areas of the brain the lexical area, it would be difficult to justify any of the Lexical Morphology Hypothesis with such evidence. This hypothesis assumes (i) that lexemes and grammatical morphemes belong to one class, the traditional isomorphic morpheme, and (ii) that both are stored in the lexicon. It can hardly be the case that the two major linguistic areas are processors, but not storage sites, since no variety of aphasia, short of global damage over both areas, presents a unitary symptom of loss of both classes simultaneously as identical storage location would predict. Either Wernicke's and Broca's areas in fact are areas of storage or they represent lexical and morphological retrieval devices with storage generalized throughout the brain, i.e. indistinct from general knowledge. Either way, theories which assume that morphemes and lexemes are of the same linguistic class are not substantiated in the neurological data as are those which distinguish the two.

Research in aphasia in the US has most recently focused on the relations of syntactic derivation disturbed by agrammatism. Agrammatism is traditionally explained as a deficiency of the central grammatical or syntactic processor (Bradley, Garrett and Zurif 1980). The basis for this conclusion is the disturbance or loss of morphemes which mark syntactic relations and the confusion of subject and object positions even when morphologically unmarked.

The most recent research in agrammatism, however, has brought doubt on the involvement of a unitary central grammatical processor (Berndt and Caramazza 1980, Kean 1985). Three reasons may be cited: (a) cases of agrammatism have been disco vered which differ in the relative inaccessibility of syntax and morphological markers, (b) the difficulty in accessing verbs, a deficit also common (and more explicable) among sensory aphasics and, finally, (c) aphasics who recognize that their sentences are ungrammatical, whose grammar thus seems independent of syntactic computation. In this section I would like to examine each of these new problems in light of the model outlined in Section 2 and show that the first two problems facing the classical theory in fact speak in favor of LMBH while the third represents no more a problem for this hypothesis than for sign based hypotheses.

Miceli, Mazzucchi, Menn and Goodglass (1983) discovered two Italian-speaking patients with the classic symptoms of agrammatism yet without indication of damage to any "central language processor". Both patients in this study shared classical symptoms: omissions of articles and prepositions and the substitution of infinitives for finite forms of the verb. The second patient in this study (Case 2) exhibited an overall error rate in these areas approximately twice that of the first (Case I) thus, on the surface it seemed that he was more severely agrammatic than the other. But closer examination revealed another pattern.

By computing the string length of utterances of the two patients, Miceli et al. concluded that mean string length in the speech of Case 2 was IO. I words while that of Case I was only 3.6 words. Moreover, Case 2 omitted no more. than three main verbs in some 70-odd clauses consisting of 600 words and his overall sample contains some 30 well-formed compound and complex sentences, ignoring morphological omissions. Miceli et al. conclude that Case I has a moderate syntactic deficit and a mild morphological deficit while Case 2 has an almost purely morphological deficit, suggesting discrete processors for syntax and morphology.

Grodzinsky (1984) and Grodzinsky, Swinney and Zurif (1985) have come to a similar conclusion. Working with two Hebrew-speaking agrammatic patients, Grodzinsky discovered that grammatical categories which are "strictly closed" are always morphologically marked on appropriate stem classes and are not marked only where null variants are permissible even though the proper grammatical function is seldom marked. (Carstairs 1981 defines a "strictly closed" category by the principle that one of its markers m u s t appear on the stem of the class it marks, e.g. the Latin (or Hebrew) nominal declension and verbal conjugation endings.) The relevant point which Grodzinsky makes, however, is that agrammatics misuse only appropriate affixes or revoweling schemes in such cases, never random phonological sequences. This suggests that agrammatic patients may retain a more or less intact set of morphemes even in the absence of knowledge of the syntactic functions which they are supposed to mark.

These two studies confirm the findings of the earlier work of Tissot, Mounin and Lhermitte (1973). In an test of 19 French-speaking agrammatic aphasics, compared with 20 sensory aphasics, Tissot et al. found a significant split between those agrammatics who preserved articles and most prepositions but not the appropriate syntactic order of lexemes and those which did pre serve word order but had severe difficulty with articles and prepositions. This led Tissot et al. also to divide their cases into syntactic and morphological agrammatics. Thus the growing evidence against the "classical" theory of agrammatism, showing that aphasia affects at least two subcomponents of the grammatical processor discriminately rather than damaging one indiscriminately, simply fulfills the predictions of SH.

(15) (a) The boy is catch . . . the boy is . . . out . . . the boy is catching out.
(15) (b) The boy is in the . . . hits the boy.⁷
(15) (c) Boy is hurting to it.
(15) (d) Hit. The man is throwing the ball.
(15) (e) The boy is catch . . . the boy is hitch . . . the boy is hit the ball.

Notice that responses (15b) and (15d-e) indicate that the problem might not be one of retrieving the correct verb so much as establishing the proper grammatical relations with the verb selected. Since the agrammatic aphasic does not have the grammatical functions subject and object available to him, i.e. cannot encode or decode any morphological marking of subject and object including word order (Caramazza and Zurif 1976), he resorts to a semantic strategy. His sentences will begin with a noun referring to the most prominent, usually animate, referential target. Other lexical items will be selected for their semantic relation to the first chosen item. If we assume that a modification of a lexeme includes establishing the lexeme's position in the phrase (word order), or extend our definition of "morpheme" to include modifications of phrases as well as lexemes, this data becomes yet another prediction of LMBH.

A more serious challenge for the classic theory is the work of Miceli, Silveri, Villa and Caramazza (1984). This study elicited verbs and nouns in isolation (not in syntactic context) from agrammatic, anomic and normal subjects and showed that (some) agrammatics in fact have a verb deficit. Citing other evidence which shows a dissociation of verb loss and syntactic processing, however, this study concludes that verb loss results from damage "to independent cognitive systems that tend to co-occur because of the neural proximity of the two systems such that when one system is impaired there is a high probability that the other will be impaired" (Miceli et al. 1984, 218). It is still possible, therefore, to explain the difficulty of strictly agrammatic aphasics in verb selection in terms of a failure of grammatical processing rather than one of lexeme selection.

Linebarger, Schwartz and Saffran (1983) and Schwartz, Linebarger and Saffran (1985) have studied the ability of agrammatic patients to restart and correct themselves, behavior which suggests that they often realize that the sentences which they are speaking are agrammatic. If this is true, agrammatic aphasics might retain intact syntactic compe tence and only suffer a deficit of the parser .This would make it possible to argue that the lexeme/morpheme distinction is an artifact of speech production which has no correlate in competence, ergo a theory of grammar.

The study by Linebarger and her colleagues did demonstrate that their four agrammatic patients could distinguish grammatical from ungrammatical sentences at a much better than chance rate. 207 of 230 grammatical sentences were recognized as such and 163 of 221 bad sentences were recognized as such on the average. This leaves 23 of 230 grammatical sentences still misperceived as ungrammatical and 58 of 221 ungrammatical sentences mistaken as grammatical. This study begs at least two comments.

First, none of the four patients in the study responded perfectly as normal subjects presumably would. They performed worst on constructions with tag questions or reflexives, i.e. those whose interpretation rests solely on the manipulation of grammatical morphemes and relations and which gain nothing from recourse to lexical or pragmatic semantics. Thus some deficit in the basic knowledge of grammar and/or morphology clearly underlies agrammatism even in the data of Linebarger et al.

Second, recognition studies corroborating greater ability at recognizing vis a vis producing grammatical structures do not prove agrammatism to be a performative deficit if the disproportion is also present in normals. The phenomenon of comprehension quantitatively far exceeding production is a general characteristic of language performance. First and second language learners at every stage of learning are capable of comprehending far more than they are able to produce. This phenomenon probably characterizes mature speakers as well; I do not know that it has been researched or explained.

It would be reasonable to predict that the tactical ability to recognize phrases as grammatical is a subcomponent of comprehension. If so, as long as this general phenomenon is maintained in approximately the same proportion and is not reversed or otherwise affected in agrammatic speech as compared to normal, it may not be a symptom of agrammatism. The interesting aspect of agrammatism is that, overall, the ability of a speaker to retrieve and produce grammatical material is grossly impaired in comparison to normal speakers and the proportion of impairment of comprehension and production remains the same, i.e. given the presumably normal superiority of speech comprehension over speech production. This set of facts embarrasses LMBH no more than traditional sign-based theories.

Up to this point we have seen how processors previously considered unitary in fact reflect discrete components. One prediction by LMBH, however, predicts the integrity of components many linguists consider discrete. The Unitary Function Hypothesis predicts that the morphology of lexical derivation and inflection is unitary and identical. The Type Transparency Hypothesis predicts that both will be computed by the same mental processor. Thus the damage to inflectional morphology witnessed in agrammatic patients should be accompanied by damage to lexical derivational morphology. The Split Morphology Hypothesis (Perlmutter 1986) predicts that lexical morphology is associated with lexical storage while inflectional morphology comprises independent postlexical operations.

Both sensory and agrammatic aphasics should exhibit lexical derivational deficits however, differences in the natures of their deficits should detectable. Agrammatic patients should have access to lexemes, but have difficulty in deploying productive affixes; their speech should exhibit derivational errors and a reduction in the rate of use of productively derived lexemes, but a retention of lexicalized (idiomatized) derived lexemes. Sensory aphasics, however, have difficulty accessing lexemes, the stems which are subject to lexical derivation their speech should exhibit productive affixes added to paraphasias or neologistic stems. In comprehension, we should find evidence that sensory aphasics resort more to lexical derivation in an attempt to recover the meaning of lexemes which they cannot comprehend while agrammatics ignore derivational opportunities and attempt to interpret lexical derivates as idiomatic lexical wholes.

One caveat demands attention in examining the evidence for the computation of derivational morphology. Since speakers have two means of accessing derived lexemes, storage and generation, the presence of a high percentage of properly marked derived lexemes in the speech of agrammatic speakers does not necessarily indicate split morphology. The prediction of the Unitary Function Hypothesis is that the morphological rules for marking productive lexical derivations will be affected by insult to the same area which controls inflectional morphology. The problem here is that "productivity" is ill defined. Moreover, productivity probably varies from speaker to speaker; that is, not every speaker will productively generate the same lexical derivates.

While generative lexical rules are available to speakers of languages, if they hear or use a lexical derivate often enough, they might just as well store it. Some speakers will derive unaccountability others, who use it more, will store it. Evidence for the Unitary Function Hypothesis is found in the work of Tissot, Monnin and Lhermitte (1973) and Dressler (1977, 1979) but most of all in that of Luria's students, Akhutina, Cvetkova, Glozman and Teplickaia. Glozman (1974), Cvetkova and Teplickaia (1975) and Cvetkova and Glozman (1978a, 1978b) found that all aphasics exhibit agrammatic speech but those diagnosed as suffering from motor (Broca's) aphasia cannot account for the compositionality of derived lexemes and treat them as units.

Patients suffering from posterior aphasias, acoustico-mnestic (sensory) and semantic in Luria's terms, rely heavily on compositional elements and tend to treat them transparently, ignoring lexical idiomaticities. Dressler (1977) found similar evidence in German-speaking patients, concluding that the "lexicon and word formation seem to be different components stored in different parts of the brain" and that the latter is a part of the same component which controls syntax.

Cvetkova and Glozman (1978a) conducted two experiments with motor and dynamic aphasics, suffering from anterior insult, and semantic and acoustico-mnestic aphasics, suffering from posterior insult. The acousticomnestic (sensory) aphasics exhibit a significantly lower rate of agrammatism than the other three types and have all the general symptoms of sensory aphasia. Semantic aphasics share the symptoms of Broca's and sensory aphasias. In the Cvetkova and Glozman experiments, all categories of aphasics were compared with a normal control group. The first test involved the analysis of sentences without spaces, e.g.

(16)(a)	nakruglomstolestoitvazasosennimicvetami
	'onacirculartablestandsavasewithfallflowers'

(16)(b)	na krug-l+om stol+e stoi+t vaz+a s osen-n+imi cvet+ami
	on circul-ar+Loc table+stand+s vase+Nom with autumn-al+InsPl flower+InsPl
	'on a circular table stands a vase with autumnal flowers'

Cvetkova and Glozman discovered that the strongly agrammatic motor and dynamic (also semantic) aphasics, on the one hand, and the marginally agrammatic sensory patients, on the other, tend to make different kinds of mistakes on this test. One class of mistakes were of the type na kruglom 'on the-circular', i stala 'and became', ja byl 'I was'. Cvetkova and Glozman calls these "lexical" errors because they overlook grammatical markers (prepositions like na 'on', conjunctions like i 'and', pronouns like ja 'I') and treat phrases as unitary lexical items, according to their semantic content. 91% of the motor aphasics' errors on the first test were of this type as compared to 18% of the sensory aphasics'.

The sensory aphasics tended to make what Cvetkova and Glozman consider "derivational" errors, for they indicate an erroneous lexeme division at morpheme boundaries, e.g. krug lom, stal a. This suggests that the sensory aphasics overgeneralize morphological analysis, perhaps not recognizing stems. 82% of the errors of the sensory aphasics were of this type while this type accounted for only 9% of the motoraphasics' errors. While the motor aphasics, presumably with the most greatly reduced morphological capacity, did not mistake affixes for lexemes, they did tend to fail to distinguish grammatical morphemes from the lexemes to which they attach or with which they form a grammatical unit. The sensory aphasics, however, presumably with a greatly impaired ability for lexical retrieval but minimally impaired access to grammatical rules, tended to isolate the lexemes by separating all grammatical morphemes from them. This overseparation is always along correct morphological lines with indifference to lexical (kruglom) and inflectional (stala) morphology.

Cvetkova and Glozman's second test was a classification test consisting of words written on 15 cards which normal and aphasics with the same four syndromes as defined by Luria, were asked to classify according to their semantic relationships. The cards contained genuine lexical derivates like let-at' 'to fly' and let-atel'nj 'flying (Adj)', veter 'wind' and vetr-jannyj 'wind(y)'), and potential but false derivates like vert-o-let 'revolve-o-fly' = 'helicopter' and vertikal' 'vertical (line)', and rubl' 'ruble' and rub-it' 'chop'.

Again two types of errors were observed again corresponding to the type of general impairment; the motor (dynamic and semantic) as opposed to sensory aphasics. The motor aphasics tended to coclassify any lexemes with formally similar stems, reading from left to right, e.g. vertolet : vertikal' , rubin 'ruby' : rubit' 'chop' significantly more than did the sensory aphasics (66% vs. 34% of their errors), and to ignore genuine lexical derivational relationships like leta-t' : leta-tel'nyj. Again, this was taken as an indication of overgeneralized lexical analysis at the expense of morphological (derivational).

The sensory aphasics' performance was just the opposite, only 20% of their errors were lexical in this sense while 80% were morphological (derivational). The sensory aphasics, for example, tended to co-classify words like namererenie 'intention' : merit' 'measure', rublennyj 'chopped' : rubl' 'ruble', where genuine morphological boundaries are detectable though synchronically irrelevant (idiomatized). The sensory aphasics tended to "reetymologize" these words, i.e. analyze them as derivationally related; they confused phonologically similar but semantically unrelated stems without morphological boundaries, e.g. rubin : rubit' , significantly less.

5. Conclusion

LMBH makes three major predictions: (a) there are two basic units of grammar, the "lexeme" and "morpheme" as described here, (b) lexical and syntactic derivation are independent of all rules of morphological marking and (c) the categories of lexical morphology (word formation) and inflection is unitary and integrated. The same categories that underlie inflection in syntax, underlie lexical derivation in the lexicon. The simplest imaginable versions of the Sign Base Hypothesis, the Lexical Morphology Hypothesis, claims that all morphology is lexical. This hypothesis makes at least two predictions as to basic lexical units and processes, (a) all units are lexical signs and (b) only one process handles them: lexical insertion. Thus the Lexical Morphology Hypothesis is far simpler and more constrained than the LMBH, shifting the burden of proof upon the latter.

But all the fundamental predictions of LMBH are borne out by neurological evidence. Moreover, the data of language acquisition and normal speech errors add further confirmation. Garrett (1980), in summing up his extensive research in speech error analysis, concludes " . . . that shift errors [e.g. I haven't satten down and writ . . .; I want to eated; I had forgot abouten that; . . . point outen] attach morphemes to stems without regard to any factor other than word boundaries and word form, in particular, without regard to the lexical identity or even grammatical category of the error site." His findings confirm similar evidence discussed by Bierwisch (1981) for SH, that the phonological formants of affixes are independent of their functions (outen cannot be in the past tense for prepositions have no tense).

Moreover, he argues that shift errors like these must occur at a processing level other than that which accounts for "word-exchange" errors, e.g. I don't know that I would hear one if I knew it). Garrett, like Bierwisch, points out that the word-exchange errors generally reflect lexically marked morphology as this example does (knew), while morpheme-shift errors apparently do not (crucial data is scarce). Errors at the point of lexical insertion and morphological realization in (7) account for Bierwisch's and Garrett's analyses of the speech error data neatly. Evidence of the Separation Hypothesis is also found in language acquisition studies. Several studies have shown that children learn the cases and their functions (case "semantics") prior to mastery of affixational rules, e.g. Bogoyavlenskii (1973) and the sources she cites.

This disjuncture is similar to the overgeneralization of productive affixes which characterize certain stages of acquisition. During these stages, children who previously used, for example, irregular and subregular past tense verb forms such as caught, hit and saw correctly, begin saying catched, hitted and seed in alternation with the grammatical, subregular forms. However, as Maratsos (1979, 312-313) points out, since both alternates are used in the apropriate grammatical situations, the use suggests a disjuncture between knowledge of form and that of the function.

This same evidence is incompatible with the LAH though not with word-and-paradigm models which separate lexical from inflectional derivations (Matthews 1972, Anderson 1982, 1984). However, word-and-paradigm models will probably have to adjust to the evidence of the Integral Morphology Hypothesis and combine inflectional and lexical morphology into one component. The second prediction of LMBH not only survives the neurological evidence, but finds support there as a secondary characteristic of linguistic behavior. The prediction that knowledge of the functional constraints on lexical as well as inflectional derivation are independent of the affixational constraints contradicts the predictions of all sign-based morphologies, Word-and-Paradigm included. All sign-based models will require heavy translational machinery in performance theory. If we accept the Type Transparency Hypothesis, therefore, Lexeme-Morpheme Base models of morphology are decidedly superior (more isomorphic with reasonable performance models) to Sign Based models.

¹The number of exceptions is remarkably small, mostly in the direction of the intransitive marker marking transitive verbs: student, assistant (but note assistant to the manager), disscussant. The few intransitive Latinate verbs not receiving the now produc tive -ee but receiving -er either end on -nt already (dissenter) or have become germanicized (reveller , cf. *revelant).

⁵The admittedly marginal objective nominalizations like fry-er, broil-er cannot be treated as subjectives derived from middle or ergative argument structures like "The chicken fried (well)" unless we wish to distinguish such forms from keep-er (fish) and loan-er (car) which fail the test; keep-er ≠ "a fish which keeps (well)".

>⁶The definitions of lexeme and morpheme given here are not to be confused with those of Martinet (1961). Martinet is also a structuralist and his "morphemes" and "lexemes" (variants of Martinet's "moneme") are traditional structuralist signs distinguished primarily by their belonging to closed vs. open classes. The definition of "morpheme" offered here and in Beard (1986) radically differs from that of "lexeme".

⁷Notice that patients (15a-c) seem to have access to the phonology of prepositions and particles ( out, in, to) without control over their functions. This sort of evidence is frequently cited in the literature in support of the separation of derivation and morphology (SH).

Dressler, W. U. and Barbaresi, L.: "How to fix interfixes: on the structure and pragmatics of Italian (and Spanish, Russian, Polish) antesuffixal interfixes and of English intermorphic elements." In: Acta Linguistica Academiae Scientiarum Hungaricae, Tomus 36 (1-4), pp. 53-67.

Garrett, M.: "The limits of accommodation: arguments for independent processing levels in sentence production." In: Fromkin, V. (ed.): Errors in linguistic performance; slips of the tongue, ear, pen, and hand. Academic Press, New York 1980.

(3)	trail-er	N trails (the feature film, an automobile)
	thrill-er	N thrills (the experiencer)
	sink-er	N sinks (weight, baseball pitch)
	sparkl-er	N sparkles

(5)	stand -ee	(Ø)	wait-ee	(for)
	escap -ee	(from)	retir-ee	(from)
	embark-ee	(on)	retir-ee	(from)
	embark-ee	(on)	resign-ee	(from)
	adapt-ee	(to)	confer-ee	(with)