Target MC’s response Type of error
2.11 A deficit at the morphology interface
Consider first of all what processes are assumed to take place in the morphological component. Recall that it is the level at which morphological realisations are assigned to categories retrieved from the substantive and functional lexica, prior to their entry into the syntactic component (since the morphological component is the interface between the lexica and the syntax). Inflectional morphology is added at this level, and derivational processes also hold here; affixation of substantive or functional morphemes to substantive or functional categories can be viewed as the application of Merge below the zero level (cf. Ackema 1999).
Inflectional morphemes, free and bound, are assumed to be lexically represented in the UG lexicon (or functional lexicon), which is an innately specified part of the language faculty. Their syntactic realisations are therefore qualitatively different from those of substantives; I assume that substantives are stored in their own lexicon, and that part o f their lexical entry specifies the nature of their syntactic realisation (the extended projection). Functional categories have categorial features, but following Tsimpli (1996) and Smith and Tsimpli (1995) I assume that other features of functional heads only receive their values at the morphological interface; the feature specification is made relevant in the syntax, where checking relationships are established as a result o f head movement (Chomsky 1995a). So for instance, there is a functional category AGR-S which is represented in the UG lexicon, and part of its representation is categorial features which identify it as the category AGR. However, those features are neutral until they reach the morphological interface, when agreement morphology is affixed to a verb. The morphological reflex of agreement on V must be reflected in the features of AGR-S, otherwise no checking relationship between AGR-S and V could be established in the syntax. I therefore assume that the morphological interface is the level at which the features of functional heads are realised, in a form accessible to the syntactic computation. This is preferable to the alternative assumption, that there are many instantiations of AGR-S in the functional lexicon, one for checking features of a plural subject, one for features of a person subject, and so on, and that Select (or whichever operation accesses lexical representations for the syntactic component) is somehow able to choose amongst the available alternatives, given that they would be categorially identical. It is more economical to assume that representations in the UG lexicon are categorially identified but that the specification of their features occurs at the
morphological interface as a reflex of being assigned a morphological realisation (as a free or bound category).
Notice that affixation is, in a sense, uni-directional: inflectional morphemes from the UG lexicon can affix to stems from either the UG or the substantive lexica, but derivational morphology (which is in the substantive lexicon) cannot affix to functional categories. This is straightforwardly accounted for on the assumption that categorial features of functional categories have no values until they reach the interface; substantive elements, including derivational morphemes, have s-selectional properties (Ouhalla 1991), and items with a neutralised feature specification cannot fulfil s- selectional requirements. Functional categories, on the other hand, have c-selectional properties, and these can be fulfilled by categories whether their featural specification is neutral or fully realised, since all they are interested in is categorial status (not meaning properties). Affixation necessarily precedes feature-specification, not because o f some arbitrary ordering of processes, but because feature-specification as envisaged above is a reflex of morphological processes, including inflectional affixation. I also assume, standardly, that morphology feeds into phonology, possibly via PF; I therefore propose that the realisations assigned to categories at the morphology interface are (a) accessible to the syntactic component for computational purposes, and (b) interpretable to PF.
Note that my hypothesis rests on the assumption of a deficit at the level of the morphological component only; syntax is assumed to be intact, as are the lexica and mechanisms of access to lexical representations; as are mappings between conceptual (meaning) representations and categories in the functional and substantive lexica. Given this situation, many aspects of language processing are assumed to be intact: but there will of course be ‘knock-on’ effects on the syntactic and interpretive components, because it is assumed that the morphology interface and the operations which hold there precede the application of syntactic operations: morphology feeds the syntax. If there is a problem for the morphological component in recognising functional categories, then it is possible that some unconscious strategy is applied to minimise the knock-on effects of such difficulties. This is what would lead to M C’s failure, on many occasions, to realise, recognise and interpret agreement morphology on V. Functional heads like AGR-S are available to MC, and sometimes he succeeds in recognising these (at the morphological interface) and assigning them to positions in the derivation (in the syntactic component).
However, at other times it seems that AGR-S is not projected, as when there is no auxiliary verb in the phrase MC is attempting to read (see section 2.8 above). This cannot be due to a syntactic failure to project the category, because we know that M C ’s ability to project functional heads is fine most of the time (as witness his ability to interpret number on nouns, for instance). But it could be due to some economy metric which results from and compensates for the deficiencies of the impaired morphological component. Based on the observation that the projection of a lexical VP will often suffice to get meaning across and to assign a rough interpretation to most parts of a clause, together with the fact that substantive verbs are more easily assigned a realisation at the deficient morphological interface than are functional categories like AGR-S, it is logical to propose that the morphological component does only as much work as is necessary for the syntax to function. If an auxiliary verb is present, then clearly a lexical VP will not suffice; this is when MC demonstrates that he can project higher functional heads in the clause. But when there is no auxiliary, a lexical VP should suffice, even though morphological reflexes of agreement and tense will go unchecked.
This interpretation of aspects of M C ’s performance is reminiscent of hypothesised deficits in the projection of functional structure above the clause in agrammatism (Ouhalla 1993, Grodzinsky 2000). Recall that Ouhalla proposes a far- reaching deficit which means that only the lexical VP is available to the agrammatic, resulting in an inability to interpret arguments appropriately and a particular difficulty with semantically reversible verbs, in both active and passive sentences - a pattern of performance which is rarely attested, as it turns out. Grodzinsky, on the other hand, proposes that agrammatics have a ‘pruned’ tree representation, which is cut off at the level of T. He follows Pollock’s (1989) porposed structure for the split-Infl, which locates an agreement projection below T, because this captures the performance o f some Hebrew agrammatics who appear to show (relatively) intact access to agreement morphology whilst their ability to utilise tense morphology is very deficient; the opposite pattern of performance (impaired agreement and intact tense) is not attested. There are well-documented empirical reasons, however, for assuming that agreement is probably above tense in the undamaged clause representation (e.g. Belletti 1990; Chomsky 1991, 1993), which led Chomsky to propose the existence of two agreement projections in the clause - the higher one (above T) for subject agreement, and a lower one for object
agreement. Since Hebrew is not a language with an object agreement paradigm, it seems strange to propose that Hebrew agrammatics have fewer difficulties with agreement than with tense because they are unable to project clausal structure beyond T, since AGR-S is precisely located in the portion of the tree which they should be unable to project. Grodzinsky’s hypothesis predicts a dissociation between preserved object agreement and deficient tense and subject agreement; but Hebrew does not provide an appropriate testing ground for this hypothesis. In any case, MC shows difficulties with both tense and agreement, and also shows that he can at times build the necessary structural representations for both categories; so his performance does not shed much light on the representational issue raised by Ouhalla and by Grodzinsky. His difficulties in the projection of functional domains really do have the flavour of some (possibly extra-linguistic) economy metric rather than a real representational deficit.
Notice that, if the morphological component is faulty, it is likely that almost every utterance produced or interpreted via the syntactic component will be one that crashes at the interfaces for some reason or another. Consider spoken language output, for instance; assuming that Select applies at the lexical interfaces with the morphological component, and that sub-zero operations which apply in the morphology are somehow damaged or deficient, then the representations which the morphological component yields to the syntactic component will be such that it is impossible (or at least difficult) for them to be processed normally by the syntax. If features associated with morphology have been scrambled or omitted, then these features cannot be checked in the syntax either, and syntactic errors will result. For instance, assuming that Case is assigned a realisation by the morphological component, but functional categories like Case cannot be recognised by a faulty morphology, then it is perfectly possible that Case will not be correctly morphologically realised, which in turn means that it cannot be checked in the syntax, uninterpretable features remain in the derivation and a crash results at the interpretive interface. Such errors, however, do not indicate that there is a problem in the syntactic component per se; they simply suggest that the syntax will process whatever is fed into it, and that an interpretation will be assigned at LF, even when there is no possibility that a grammatical derivation can be yielded by the computation. This is compatible with Chomsky’s (1986) view, that the language faculty processes any kind of input which consists of matched sound-meaning pairs. On the other hand, it is likely that
syntactic processes will provide some support for the deficient morphological component, so that categories which cannot be recognised at the level of the morphological component could to some extent be specified by the application of category-sensitive syntactic processes.
Notice that this view leads directly to the prediction of a contrast between the production of certain items in a sentential (syntactic) context, and the production of the same items in isolation, without syntactic support. Items which do not have a full independent lexical representation, and which therefore would normally require a morphological realisation to be assigned to them at the interface, are presumably those involving affixation (inflectional or derivational) and functional categories. Monomorphemic substantives have a full lexical representation, including information about their syntactic representation (in the sense of Grimshaw’s 1991 ‘extended projection’), when they arrive at the morphological interface via Selection. This could be a source of support for a faulty morphology component which cannot appropriately assign realisations to functional categories in isolation; if they are associated with a substantive which has an extended projection, then they can be featurally specified by means of that association.
Let us attempt to be more concrete, considering first how the effects of a deficit at the morphological component could be manifested in the language production of an aphasie patient. The most obvious effects will be on derived and inflected words. First of all, assuming that the deficit is less than total, performance is expected to be variable and to include some (rare) correct productions of affixed words. Errors are unlikely to be consistent, such that all words affixed with the nominalizing -ness are produced as if they are affixed with the adverbializing -ly, for instance; this pattern >of performance would indicate a deficit in the lexical representation of bound morphemes, and we are assuming that lexical representations have been spared (predicting, for instance, good performance on lexical decision tasks involving both function words and substantives). Rather, it is expected that a deficit at the morphological component will result in the misidentification, substitution and / or omission of affixes. Note that principles o f UG will still hold, so that a well-formedness metric will apply to the possible combinations o f stem and affix - bound morphemes should not appear by themselves, and omission of affixes should only be possible when the stem is morphologically well-formed with or
without affixation. Substantive stems should be correctly read, since the deficit is one of affixation and not whole word substitution. Dlegal affixation, say of tensè morphology to a noun, is ruled out, because categorial information about free and bound categories from both lexica, including c- and s-selectional requirements, is assumed to be intact; but inaccurate affixation, say of the present tense -ing to V instead of the past tense -ed, is not ruled out, because both alternatives fulfil the s-selectional requirements of V and both are possible realisations of T.
There is predicted to be a differentially adverse effect on function word reading compared to reading of substantives, because (monomorphemic) substantives do not rely on the morphological component to receive a syntactic characterisation, whereas functional categories do. But notice that the kind of errors predicted to occur on functional categories will be different from the kind of errors predicted for morphologically complex substantives. In the latter case, the morphological component will provide a legal realisation which may or may not reflect the target; the stem will be correctly identified. In the former case, however, the morphological component is unable to recognise the functional item in isolation, because its categorial features are neutral and it is not able to recognise them. This will yield a morphophonological representation o f some underspecified functional category, which I hypothesise will be assigned random (and possibly multiple) possible realisations by recourse to the phonological features of lexical representations in the UG lexicon.
It should be becoming clear that the pattern of performance I am predicting as the result of a partial deficit at the level of the morphological component is similar in many respects to that demonstrated by MC. The poor production of function words in isolation, coupled with their comparatively good production in sentential contexts; the misidentification of derivational and inflectional morphemes; the absence of illegal stem+affix combinations; the correct realisation of stems; all these things are straightforwardly predicted on the basis that there is a deficit at the level of an independent morphological component in the grammar. The properties of that component which I have outlined are simple and plausible, and for the most part have already been put forward as necessary properties of a morphological component in studies of normal (Tsimpli 1996) and abnormal (Smith and Tsimpli 1995) language and language acquisition.
Consider also pseudoword reading. MC, in common with all phonological dyslexies, cannot read pseudowords, making instead lexicalisation errors (section 2.5.1.3 above). It is very possible that there exists in the language faculty a specialised route which applies rules to orthographic representations to allow them to be read aloud, by a process of grapheme to phoneme conversion (the GPC route). This is the means by which pseudowords and real but unfamiliar words are read. There is strong evidence for the existence of such a reading route, including double dissociations from cases of acquired dyslexia (e.g. phonological dyslexies cannot use this route at all; dyslexies who fall into another syndrome category, so-called surface dyslexies, can only use this route, which leads them to produce régularisation errors on real words - reading pin t as / p i n t / , for instance. See Newcombe and Marshall 1985 for a description of surface dyslexia). So it may be the case that M C’s difficulties in pseudoword reading are due to the total loss of the GPC.
However, some interesting experimental results are suggestive in this regard. W hen given pseudowords to read which were derived by changing one letter o f a function word (the same items as he was asked to identify in the lexical decision task reported above in section 2.7), MC made some errors which suggested that he was simply searching the substantive and functional lexica for a similar letter string to which he could assign a pronunciation, as would be expected if some GPC route has become defunct. Specifically, he seemed to recognise the category of the pseudoword (if there could be such a thing), and he made errors accordingly. So he read absodutely as ‘before’, beyonz as ‘because’, whecher as ‘before, no, after’ (full list reported in Appendix VI). This seems to show that MC was identifying the pseudo word as a function word, and then making function word errors. However, not all of his errors followed this form. On some pseudowords he clearly demonstrated that he understood that the words looked like real words but were not. He read towargs as ‘BEEfore ’, with a strong emphasis on the first syllable, and commented ‘that’s strange’. He read howefer as ‘because - well, zbikA lje/, or something like th a t’ - and some of his other errors on reading
such items also demonstrate awareness of their strangeness (see Appendix VI). This strongly suggests that MC is not simply accessing a visually similar word to produce from the lexicon; rather, he seems to be struggling to assign these strange orthographic strings some realisable morpho(phono)logical representation. This is not an attempt to
dispute the existence of the GPC route for non-lexical reading, nor the likelihood that it is severely damaged in certain kinds of acquired dyslexia; rather, it seems that MC shows some evidence of attempting to read pseudowords via the morphological component which also assigns morph(phono)logical representations to real words. Since I am hypothesising that there is a deficit at this level in any case, it may possibly turn out to be unnecessary to posit a further deficit at a different level of processing. It is true, however, that many of M C’s errors appear to be typical visual (rather than morphological) errors, and it is not always possible to tell the difference between real mrophological errors and those which result from some other feature o f the target word - visual or phonological features (recall Badecker and Caramazza’s (1998) caveats to this effect, discussed in 1.34 above) - so it may be that there is an additional impairment at