As we saw above, the more limitations there are on the bases available to a lexeme formation process, the less productive it will be. In this section, we will explore different kinds of restrictions that may apply to lexeme formation processes.

We have actually looked at some such restrictions when we learned how to write lexeme formation rules. We learned that there could be different sorts of restrictions on what sorts of base an affix might attach to, including:

categorial restrictions: Almost all affixes are restricted to bases of specific categories. For example, -ity and -ness attach to adjectives, -ize attaches to nouns and adjectives, or un- attaches to adjectives or verbs.

phonological restrictions: Sometimes affixes will attach only to bases that fit certain phonological patterns. For example, -ize prefers nouns and adjectives that consist of two or more syllables, where the final syllable does not bear primary stress. The suffix -en, which forms verbs from adjectives, attaches only to bases that end in obstruents (stops, fricatives, and affricates). So we can get darken, brighten, and deafen but *slimmen and *tallen, which end in sonorant consonants, are impossible.

the meaning of the base: For example, negative un- prefers bases that are not themselves negative in meaning. We find unlovely but not *unugly, unhappy but not *unsad.

To these sorts of restrictions we might add:

etymological restrictions: Some affixes are restricted to particular subclasses of bases. For example, there are affixes in English that prefer to attach to bases that are native – for example the suffix -en that forms adjectives from nouns (wooden, waxen but not *metalen or *carbonen). On the other hand, another suffix -ic that forms adjectives from nouns (parasitic, dramatic) will not attach to native bases, only to bases that are borrowed into English from French or Latin.

syntactic restrictions: Sometimes affixes are sensitive to syntactic properties of their bases. For example, the suffix -able generally attaches to transitive verbs, specifically verbs that can be passivized. So from the transitive verb love we can get loveable, but from the intransitive verb snore there is no *snorable.

pragmatic restrictions: Bauer (2001: 135) gives the following example. In Dyirbal, there is a suffix -ginay that means ‘covered with’. Although there might conceivably be a use for a word meaning something like ‘covered with honey’, in fact, the suffix occurs in Dyirbal only on bases that denote things that are “dirty or unpleasant” (Dixon 1972: 223), like gunaginay, which means ‘covered with feces’. What’s considered dirty or unpleasant might to some extent be a function of cultural beliefs.

We might expect there to be an inverse correlation between the number of restrictions and the productivity of a lexeme formation process: the more restrictions apply, the fewer bases it will have available to it, and the fewer words it will be able to derive.

The restrictions above pertain to inputs to lexeme formation rules. But it’s also possible for there to be restrictions specifically on the output of rules. For example, certain sorts of complex words can be restricted in register. Baayen (1989: 24–5) notes that the suffix -erd in Dutch forms “jocular and often slightly pejorative personal names.” For example, from the adjective bang ‘afraid’ we get bangerd ‘fraidy-cat’ and from dik ‘fat’ we get dikkerd ‘fatty’. Baayen points out that although there are a lot of adjectives that might give rise to pejorative names for people, words formed with the suffix are confined to use in spoken, as opposed to written, language and therefore the output of this lexeme formation process is restricted.