Crowdsourcing and Minority Languages: The Case of Galician Inflected Infinitives.

Results from  a crowdsourced audio questionnaire show that inflected infinitives in Galician are acceptable in a broad range of contexts, different from those described for European Portuguese. Crucially, inflected infinitives with referential subjects are widely accepted only inside strong islands in Galician (complements of nouns, adjunct clauses). They are widely rejected in non-islands, notably in the complements of epistemic/factive verbs, in contrast with Portuguese and older varieties of Galician (Gondar, 1978; Raposo, 1987). Statistical analysis shows, however, that, in the complements of epistemic/factive (and desiderative) verbs, inflected infinitives are significantly more acceptable in instances of control, whether partial or exhaustive. In fact, there is no significant difference between these two types of control in Galician, unlike in Portuguese, where inflection is generally better in instances of partial control and is not acceptable in instances of exhaustive local subject control (Modesto, 2010; Sheehan, 2018a). We propose an analysis of this pattern in terms of phase theory. The inflectional domain of non-finite clauses remains visible to the thematic domain of the next clause up, according to the less strict version of the Phase Impenetrability Condition (Chomsky, 2001), allowing control to take place. Pronouns/or pronominal inflections in the inflectional domain of visible non-finite clauses therefore get controlled. In islands, however, material in the inflectional domain remains free/referential. Despite this basic pattern, the data are characterized by substantial interspeaker variation. Statistical analysis shows that gender, urban/rural birthplace and mother tongue are all significant factors in this variation, while age and region of birth are not. Most notably, urban-born male bilinguals with Spanish as their mother tongue consistently rate all sentences higher on the Likert scale. Overall, the results show that crowdsourcing can lead to empirically robust syntactic descriptions of minority languages which are likely to be subject to substantial sociolinguistic variation and where judgments from a single social group may be misrepresentative of the general picture. The study also highlights, however, the challenges associated with using crowdsourced audio-questionnaires of this kind and the need for statistical analysis of results to control for substantial amounts of variation.