Construction morphology (review)

In his latest book Geert Booij presents a construction-based theory of morphology in which a number of morphological phenomena, primarily taken from Dutch, are analyzed according to the principles and practices of construction grammar and related work, such as head-driven phrase structure grammar (HPSG), cognitive linguistics, and Culicover and Jackendoff’s (2005) Simpler syntax approach. The emphasis is on the exposition and motivation of this approach, and there is little in the way of polemics against other approaches, such as the distributed morphology framework (cf. Halle & Marantz 1993) that is popular among minimalists.

Among the advantages of the construction-based approach, Booij lists the fact that is it not input-oriented in the way that rule-based accounts are. For instance, a rule of -ist affixation might seem appropriate to deal with pairs such as Marx – Marxist, and another rule of -ism affixation for the pair Marx – Marxism, but, as was pointed out by Aronoff (1976), the relation between such pairs as Marx and Marxist or that between Marx and Marxism is linguistically opaque (depending on the piece of world knowledge that Karl Marx instigated a political movement and not, say, a speech error—cf. spoonerism, named after the Reverend Spooner), and the only clear semantic relationship we can establish is that between Marxism and Marxist. Affixation rules run into problems with cases like baptist – baptism, given that the element bapt does not seem to be a meaningful morpheme, and yet both baptist and baptism have a clear meaning. For Booij, the solution is to posit a ‘paradigmatic relationship’ defined over two constructional schemas.

We may read this as: a word in -ism, meaning SEMi is paradigmatically related to a word in -ist denoting people with some unspecified property related to SEMi. These paradigmatic relationships have a status somewhat like the rules of yore: they have to be learned (by children) and form part of the specification of the grammar. Not every relationship between words is grammatically relevant. For example, synonymy is a relationship between words, just like the one in 1, yet presumably it is not part of the grammar. A difference with rules is that the relationships have no stipulated direction. Example 1 does not state that words in -ist are derived from words in -ism or vice versa. This may appear to be a welcome conclusion, since there are words in -ism lacking a counterpart in -ist (a filmmaker who indulges in Hitchcockisms is not called a Hitchcockist) and words in -ist lacking counterparts in -ism (e.g. dentist). I submit that this property of nondirectionality, however, can also be seen as a disadvantage. It is commonly argued, to my mind correctly, that there is a basic difference between adding an affix and subtracting one, a difference that is captured by assuming a directional rule of affixation, and by delegating affix subtraction to a process of backformation that is supposed to have a different status in the language. As Booij notes (26), a new verb such as skype may right away give rise to an agent noun, skyper. The opposite does not appear to be true, however. Somebody hearing the word rudder ‘instrument for steering’ won’t be as free to postulate a verb rud ‘to steer’ on the basis of the noun. Indeed, the corresponding verb in English is to rudder, using transposition (a productive process) rather than affix subtraction. Although one may find a fair amount of cases of backformation in languages such as English or Dutch, their numbers pale in comparison to cases of regular affixation. This difference in productivity is not captured by the static and symmetrical lexical relations that take center stage in Booij’s account.

In his discussion of exocentric compounds, like English pickpocket, Booij notes that one might analyze these elements by postulating a zero affix associated with the semantic component ‘agent/instrument’ (like English -er), because that part of the meaning of these compounds cannot straightforwardly be predicted from the meanings of the component elements pick and pocket. Rather than postulating a zero...