Semantic features in language comprehension: Why 'cattle' can be primed by 'furniture'

Heike Wiese, Humboldt-University Berlin and Maria M. Pinango, Yale University

Traditionally, semantic and conceptual representations have not been distinguished for the purposes of comprehension. Current research on the syntax-semantics interface, however, suggests the possibility of a semantic level of meaning, SEM, with a linguistic organisation that mediates the correlation of syntactic and conceptual representations. We investigate the existence of such a level of representation as comprehension unfolds. Consider the following triad: 'cattle', 'cows', 'beef'. 'Cattle' falls into the same class as 'beef' morpho-syntactically, because both are transnumeral (ie. neither pluralises), but falls into the same class as 'cows' conceptually, since both denote objects, rather than substances.
Hence morpho-syntactic and conceptual classifications are not in a one-to-one correspondence.

The correlation between those classifications is mediated on the semantic level by two features, [struc] and [ind]: semantic representations are [+struc] if they have an internal structure that determines what counts as a minimal instance of the nominal concept; they are [+ind] if they contain an individuation function that provides direct access to individual elements. These features distinguish 'cattle' from both 'cows' and 'beef' on the semantic level: the semantic representation of 'cattle' differs from that of 'cows' because it does not contain an individuation function ('cattle' is [-ind], whereas 'cows' is [+ind]), and it differs from that of 'beef' in having an internal structure ('cattle' is [+struc], whereas 'beef' is [-struc]). These semantic features identify three nominal classes:

[+struc, -ind]: collectives (eg. 'cattle' or 'furniture')
[+struc, +ind]: plurals (eg. 'cows')
[-struc, -ind]: mass nouns (eg. 'beef').

We hypothesise that if this level of representation is accessed for the purposes of comprehension, priming effects based on these properties should be observed. This predicts that collectives such as 'furniture' should have a facilitating effect for other collectives such as 'cattle' (coll-coll), as compared to plurals like 'cows' (coll-plu) or mass nouns such as 'beef'
(coll-mass).

Previously we showed that in English 'collectivity' has a facilitating effect with respect to nominals [-struc, -ind]. In follow-up experiments in German, using a lexical decision design, we have now replicated these results and shown that our prediction is borne out for the combination of [struc, ind] (mean[coll-coll] = 710.85ms < mean[coll-plu] = 733.17ms, p =
0.0048; mean[coll-coll] = 710.85ms < mean[coll-mass] = 734.88, p = 0.008).

These results thus strongly support the existence of SEM as a level of representation, suggesting that SEM features are accessed as comprehension unfolds.