Are non-semantic morphological effects incompatible with a distributed connectionist approach to lexical processing?

Abstract

On a distributed connectionist approach, morphology reflects a learned sensitivity to the systematic relationships among the surface forms of words and their meanings. Performance on lexical tasks should thus exhibit graded effects of both semantic and formal similarity. Although there is evidence for such effects, there are also demonstrations of morphological effects in the absence of semantic similarity (when formal similarity is controlled) in morphologically rich languages like Hebrew. Such findings are typically interpreted as being problematic for the connectionist account. To evaluate whether this interpretation is valid, we carried out simulations in which a set of morphologically related words varying in semantic transparency were embedded in either a morphologically rich or impoverished artificial language. We found that morphological priming increased with degree of semantic transparency in both languages. Critically, priming extended to semantically opaque items in the morphologically rich language (consistent with findings in Hebrew) but not in the impoverished language (consistent with findings in English). Such priming arises because the processing of all items, including opaque forms, is influenced by the degree of morphological organisation of the entire system. These findings suggest that, rather than being challenged by the occurrence of non-semantic morphological effects in morphologically rich languages, the connectionist approach may provide an explanation for the cross-linguistic differences in the occurrence of these effects.