Learning to lose: the role of input variability in the loss of V2

Abstract

The loss of verb second (V2) in languages like English and French is often attributed to language-specific causes. One feature that is shared by all languages that have lost V2 is the increase in subjects in clause-initial position. This thesis explores how the variability in constituent types occurring in the clause-initial position affects the acquisition of V2. It is hypothesised that lower variability in the input affects the acquisition process negatively: Learners are unable to dissociate the clause-initial position from a specific constituent type, in particular subjects. The hypothesis is formulated against the backdrop of a domain-general fostering effect of variability on learning (cf. Raviv, Lupyan & Green 2022). The cognitive perspective that this approach takes allows the provision of a more uniform explanation for the changes that have occurred in individual languages.

At the outset, variability in grammatical functions (e.g. subject, object) was stipulated to be the relevant domain. The hypothesis was then investigated with artificial language learning (ALL) experiments and a corpus study. Building on previous work, an ALL design was conceived that enables participants to sufficiently learn a V2 grammar in a short period of time (experiment 1 & 2). This design was put to use in three experiments with native speakers of English as participants. All three experiments compared languages with different distributions of clause-initial constituents: a high-variable language (i.e. with a uniform distribution of subjects, objects and adjuncts in clause-initial position), and two or three low-variable languages (i.e. with one constituent type dominating the clause-initial position). The experiments differed in the nature of the artificial language: a semi-artificial language combining English vocabulary with a V2 grammar (experiment 3), a fully artificial language (experiment 4) and a visual language that uses icons in lieu of lexical items (experiment 5) were used. The learning success in general was conditioned on the lexicon size such that a sufficiently sized lexicon was necessary for successful acquisition (experiment 4). Moreover, the results suggest that input variability indeed affects learning of a V2 grammar. Across experiments, a language where objects dominate the clause-initial position was the most difficult to learn.

A language in which adjuncts dominate was learnt best in experiment 3, whereas no difference between such a language and a language with a uniform distribution of clause-initial elements was found for experiment 5. Although grammatical functions were initially defined as the relevant domain of variability, this could not be sustained in the light of the experimental findings. Instead, it is argued that learners can be sensitive to variability in grammatical functions and grammatical categories (e.g. NP, AdvP, CP). The uniform language exhibits high variability in terms of grammatical functions, while the adjunct-dominant language is characterised by high variability in grammatical categories. The advantage of the adjunct-dominant language in experiment 3 can be attributed to influence of participants’ L1 (English). This overall interpretation receives independent support from a large-scale corpus study showing a large proportion of adjuncts in initial position in German. Adjuncts entail high variability in grammatical functions and grammatical categories. The stability of the V2 property in German for over a millennium can be interpreted as direct consequence of a high proportion of clause-initial adjuncts. Taken together, the findings of this thesis provide support for the role the low variability in clause-initial constituent types played in the loss of V2. It further shows that a cognitive angle can be useful for finding uniform factors contributing to the loss of V2.