Statistics and semantics in the acquisition of Spanish word order: Testing two accounts of the retreat from locative overgeneralization errors DOI 10.1515/lingvan-2015-0021 Received September 1, 2015 accepted April 29, 2016 Abstract: Native speakers of Spanish (children aged 6-7, 10-11 and adults) rated grammatical and ungrammatical ground-and figure-locative sentences with high frequency, low frequency and novel verbs (e. g., Lisa llenó/forró/nupó la caja con papel; *Lisa llenó/forró/nupó papel en la caja, 'Lisa filled/ lined/nupped the box with paper'; 'Lisa filled/lined/nupped paper into the box') using a 5-point scale.Echoing the findings of a previous English study (a language with some important syntactic differences relevant to the locative), participants rated errors as least acceptable with high frequency verbs, more acceptable with low frequency verbs, and most acceptable with novel verbs, suggesting that learners retreat from error using statistically-based learning mechanisms regardless of the target language. In support of the semantic verb class hypothesis, adults showed evidence of using the meanings assigned to novel verbs to determine the locative constructions in which they can and cannot appear. However, unlike in the previous English study, the child groups did not. We conclude that the more flexible word order exhibited by Spanish, as compared to English, may make these types of regularities more difficult to discern.Keywords: child language acquisition, Spanish, locatives, argument structure overgeneralization errors, verb semantics, statistical learning, entrenchment, pre-emption Children face a paradox when they learn their target language/s (Braine 1971;Baker 1979;Bowerman 1988;Pinker 1989). On the one hand, they must form generalizations that allow them to use verbs in argument structure constructions in which they have not been encountered in the input. Constructions involving figure and ground locatives are an interesting example. For instance, in a language like Spanish, on the basis of hearing figure-locatives (1) and ground-locatives (2) with the same verb, children could set up a productive generalization that allows them to use in the latter construction verbs that have been attested solely in the former construction, and vice versa (3). Indeed, without this ability, language would consist of nothing more than an inventory of rote-learned utterances; a position that has not been taken seriously since Chomsky's (1959) review of Skinner's (1957) Verbal Behavior.On the other hand, in order to avoid producing utterances that adult speakers would regard as ungrammatical, children must somehow learn to restrict these generalizations. For example, over-