Sentences such as Three girls are holding two balloons, whose subjects and objects are quantified by bare numeral expressions, logically allow multiple readings. The semantics literature has reported that the so-called object wide scope distributive reading, (interpreted as having 6 girls and 2 balloons, each balloon held by 3 girls), is usually not accessible. Recent experimental studies showed the reading was accessible, albeit massively dispreferred (Musolino, 2009; Syrett & Musolino, in prep.) We report the findings of two experiments that tested competing theoretical accounts of why this reading should be disallowed. On one account, the syntactic configurations and operations required to generate the reading are not permitted by the grammar (Beghelli & Stowell, 1997). On the other, the reading is taken to be allowed by the grammar but rendered inaccessible by excessive processing costs (Reinhart, 2006). Crucially, this account involves the semantic nature of the subject; when it allows both a distributive and a collective interpretation, the computation of all possible readings exceeds working memory capacity. It straightforwardly predicts that if a collective reading can be forced by adding a modifier, e.g., Three girls together are holding two balloons, the object wide scope reading should become acceptable. This was reported to be the case for a small number of informally consulted subjects.

Experiment 1 piloted 5 lexical items within subjects, all with singular indefinite subjects (N = 42). Results revealed that participants accepted the object wide scope reading, as predicted. However, clear item effects were found, contra the literature. Experiment 2 varied 3 types of subject noun phrase between groups: singular indefinites, bare numeral quantifiers and bare numeral quantifiers plus a collectivizing modifier. It also tested 4 lexical items from Experiment 1 within subjects (N = 132). Results revealed: 1- the object wide scope reading was acceptable to most participants, with or without modification, contra both theoretical claims and previous experimental findings; and 2- clear differences among lexical items, replicating Experiment 1. These findings suggest that participants did not treat all internal arguments equally. Rather, they were sensitive to the argument structure of the verbs, distinguishing true transitives from unaccusatives.