The word formation domain is a linguistic field where interfaces between different structures of language are activated. The construction of lexemes depends on the semantic, syntactic and phonological features of the bases and affixes involved in the process. Derivation, as a kind of word formation process, obeys selectional restrictions (Plag 1999, 2004) that intertwine in interfaces (Rodrigues 2008). Inside those semantic, syntactic and phonological features, many substructures or tiers become involved in the choice between bases and affixes.
As is well known, there are different affixes that operate in the same word formation rule (henceforth WFR) by adjoining the same syntactic bases and producing the same categorial and semantic derivatives. For instance, in Portuguese there are many affixes that attach to verbs and create deverbal event nouns. Amongst those affixes are, for example, -ção, -mento, -aria, -dura, -ão, -nça. However, it is not possible to attach all of these affixes to every verb base (Rodrigues 2008). Morphological, phonological, semantic and processing constraints intervene in the choice of the affix. Selectional restrictions are, thus, in interface with each other and with the processing conditions (Hay 2002, Hay and Plag 2004, Plag and Baayen 2009, Rodrigues 2008, 2009).
The problem of selectional restrictions arises when the same affix operates in more than one WFR. Depending on the WFR it is working on, the affix imposes different selectional restrictions on the bases.
This is the case of the affix -ão, which operates in Portuguese agent and event deverbal formations. The behaviour of this affix raises problems for the conceptualisation of WFRs, of the affixes’ roles, and even of the status of selectional restrictions. To solve those problems, we have proposed a new model on word formation. This model is called ‘WFRs in interfaces’ (WFRIM) (Rodrigues 2008).
According to this model, affixes should be considered neither as a simple phonological spell-out of WFRs, as proposed by separationist process-oriented approaches (e.g. Beard 1995), nor as being totally responsible for the construction of derivatives, without the alignment of a WFR, as defended by output-oriented models (e.g. Plag 1999). In addition, selectional restrictions and processing conditions are not the only constraints responsible for the grammatical and ungrammatical attachments of affixes to bases. The maximal semantic compatibility between the features of the affix, those of the base and of the WFR, will determine which interfaces will occur between affix and base. In this way, some meanings will be achieved whilst others will not, in one WFR, and the opposite will happen in another WFR.
