Structural restrictions: Rule-specific mechanisms

Structural restrictions in word-formation may concern the traditional areas of linguistic research, phonology, morphology, semantics and syntax. A general question that arises from the study of such restrictions is which of these should be considered to be peculiar to the individual rule and which ones are the consequence of a more general mechanism (or restriction) that operates on all (or at least classes of) morphological processes, a question that can only be answered on empirical grounds. The first type of structural restrictions we will discuss here are restrictions that are only operative with a specific process and do not constrain derivational morphology in a principled way. The general constraints will be discussed in Structural restrictions: General mechanisms.

Rule-specific constraints may concern the properties of the base or the derived word (see e. g. Bauer 1982). Consider, for example, the English personal noun suffix -ee and the adjectival suffix -able, both of which are often claimed to attach only to transitive verbs.¹ The theoretical status of such syntactic restrictions has, however, been questioned on the grounds that the syntactic specifications are in fact the result of underlying semantic restrictions. In fact, I will argue for this position in Rival morphological processes 2: The structural properties of -ize derivatives (see also the discussion of the unitary base hypothesis below). An example of a semantic restriction proper is the one imposed on -ee derivatives, whose denotata must be sentient entities (e.g. Barker 1995).

Rule-specific morphological constraints make reference to the kinds of morphological elements or features (such as [+ Latinate], see below) involved in a process, for example by ruling out certain combinations of affixes, or by making affixation dependent on the presence of a particular affix in the base. The nominalizing suffix -ity is a case in point, because it may not be attached to adjectives ending in -ory (cf. * satisfactorily). Another affixation process making reference to the non-presence of a particular affix on the base is the German perfect prefix ge- which is not attached to stems that feature verbal prefixes such as be-, er-, ver- and the like (cf. *gebesprochen, *geerblindet, *geverarztet, Carstairs-McCarthy 1993, but see below for an alternative solution).

Phonological constraints may involve segmental and prosodic restrictions. For example, the deverbal nominalization suffix -al occurs only on stress-final verbs like recite or propóse and verbal -en only attaches to monosyllabic bases that end in an obstruent (e.g. Siegel 1971, Gussmann 1987). The full complexity of phonological constraints will become evident when we consider the segmental and prosodic restrictions on derived verbs.

A number of interesting empirical and theoretical problems emerge with rule-specific restrictions. As already mentioned above, it is often not clear whether the restriction is indeed rule-specific or the consequence of a more general restriction. A notorious case is haplology, i.e. the avoidance of similar-sounding sequences within and across words. Standardly, cases of haplology have been formulated as constraints on specific rules,² but given the proliferance of similar phenomena and the similarities between individual instances of haplology, a more general solution would be desirable.³

Another problem is to exactly determine the provenance of the restrictions. Thus it is often uncertain whether one deals with a morphological or with a phonological phenomenon. For example, it has been argued that the apparently morphological restriction on German past participle formation mentioned above is in fact a phonological restriction. Wiese (1996a:89-98) and Neef (1996:233-242) suggest that the distribution of ge- can be accounted for by purely phonological mechanisms that make reference to the prosodic structure of the base word or the derivative. Although the two authors differ in their explanation of this particular restriction on German past participles they both observe that past participles must not bear initial primary stress, and that only in cases where this is not the case anyway (as with the prefix-verbs mentioned), ge- is prefixed. In this formulation, reference to the morphology of the base is superfluous. German past participles illustrate a more principled theoretical debate about the question whether the morphological structure of a word is "visible" to further morphological operations, a question we will not pursue any further.⁴

The third problem concerns the location of the restrictions. Very often, these have been defined as conditions that affixes impose on possible bases, i.e. as restrictions on the input to morphological rules. The concept of input-oriented constraints implies that it is the affix that selects its base and not vice versa. Although this is a widely-shared assumption, it seems that this is not a logical necessity, so that there appears to be no principled argument against the view that it is the base (with its affixes) that constrains the selection of possible affixes. We will indeed argue that such base-driven mechanisms can have important advantages over affix-driven ones.

The adequacy of input-oriented restrictions has been further questioned on the grounds of examples like the above-mentioned verbal suffix -en, whose phonological properties should rather be defined as a restriction on the output, and not, as originally stated, as a restriction on the input of the rule. Siegel (1971) observes that -en must be preceded by one and only one obstruent, which may be preceded by an optional sonorant preceded by a vowel (cf. also Gussmann 1987). This generalization runs into difficulties if formulated as a restriction on the input, since derivatives like soften and fasten would be ill-formed because their base words end in two, and not one, obstruent. However, if formulated as a condition on -en derivatives, the derived verbs are well-formed because only one obstruent, and not two, precedes the suffix. I will argue in detail that the phonological restrictions on derived verbs are best accounted for in terms of output restrictions.

In summary it is clear that, irrespective of the empirical and theoretical problems in finding the correct generalizations, rule-specific restrictions play a prominent role in word-formation and should be integrated into the formalization of any derivational mechanism (cf. e.g. Anderson 1992: 196).

¹ But see the discussion in Bauer (1983) and Barker (1995) who show that occasionally intransitive verbs occur as bases.

² See, for example, Booij (1983:257). A recent rule-specific approach to the well-known problem of haplology with plural -s and possessive -s in English can be found in Yip (1996).

³ Yip (1996), Plag (1998) are attempts in this direction.

⁴ See for example Kiparsky's 'bracket erasure convention' (1982a), Anderson (1992) for arguments against visibility, Carstairs-McCarthy (1993), Booij (1996) for arguments in favor of visibility.