A multimodel inference approach to categorical variant choice: construction, priming and frequency effects on the choice between full and contracted forms of am, are and is

Abstract

A multimodel inference approach to categorical variant choice: construction, priming and frequency effects on the choice between full and contracted forms of am, are and is Abstract: The present paper presents a multimodel inference approach to lin guistic variation, expanding on prior work by Kuperman and Bresnan (2012). We argue that corpus data often present the analyst with high model selection uncer tainty. This uncertainty is inevitable given that language is highly redundant: ev ery feature is predictable from multiple other features. However, uncertainty in volved in model selection is ignored by the standard method of selecting the single best model and inferring the effects of the predictors under the assumption that the best model is true. Multimodel inference avoids committing to a single model. Rather, we make predictions based on the entire set of plausible models, with contributions of models weighted by the models' predictive value. We argue that multimodel inference is superior to model selection for both the ILanguage goal of inferring the mental grammars that generated the corpus, and the ELanguage goal of predicting characteristics of future speech samples from the community represented by the corpus. Applying multimodel inference to the classic problem of English auxiliary contraction, we show that the choice between multimodel inference and model selection matters in practice: the best model may contain predictors that are not significant when the full set of plau sible models is considered, and may omit predictors that are significant consid ering the full set of models. We also contribute to the study of English auxiliary contraction. We document the effects of priming, contextual predictability, and specific syntactic constructions and provide evidence against effects of phono logical context.

Machine Meets Man: Evaluating the Psychological Reality of Corpus-based Probabilistic Models

“…Any given feature seems predictable from many other features. Because of this redundancy, an utterance can be produced in (unobservably) different ways, which explains how individual differences and uniformity across the community can co-exist (Hurford 2000;Barth and Kapatsinski 2014;Dąbrowska 2013Dąbrowska , 2014. Thus, while multicollinearity can be a major headache for statistical modelling (but see Harrell 2001), it may be a blessing for language learners, in that it enables speakers to behave in a way that is broadly similar to that of other speakers even when they all have different underlying grammars.…”

“…This, combined with the considerable differences in the performance of human participants, suggests that rather than trying to find the single "best" model, it may be more productive to develop a range of models reflecting the range of human performance (as already suggested by Lauri Carlson, cf. Arppe 2008: 208); for a practical implementation, see Barth and Kapatsinski 2014).…”

Effects of average and specific context probability on reduction of function wordsBEandHAVE

“…The few studies examining function words show that they are less affected by repeated mention (Bell et al 2009) and more affected by predictability given the preceding context (Jurafsky et al 2001), but both content and function words reduce in cases of high following context probability (Bell et al 2009). Studies of contraction of the function words HAVE and BE indicate that contraction is more likely when BE or HAVE is highly probable given the context (Krug 1998;Frank and Jaeger 2008;Bresnan and Spencer 2016;Barth and Kapatsinski 2017). In all its meanings BE is a function word, allowing a controlled look at expression differences among grammatical meanings of the same word.…”

Robust, Causal, and Incremental Approaches to Investigating Linguistic Adaptation

“…Woodward ( 2006 ) and Hoover and Perez ( 2004 ) note that ensuring structural robustness is a hard problem, especially since results may be sensitive not just to the set of control variables, but to the particular combination of control variables, causing an exponential explosion of possible control models. Barth and Kapatsinski ( 2017 ) suggest that this is a real problem for linguistics because aspects of language are highly redundant and inter-related. Instead of committing to one “best” model for the final results, Barth and Kapatsinski ( 2017 ) suggest a “multimodel inference” approach, which assesses the hypothesized relationship in a wide range of models.…”

“…Barth and Kapatsinski ( 2017 ) suggest that this is a real problem for linguistics because aspects of language are highly redundant and inter-related. Instead of committing to one “best” model for the final results, Barth and Kapatsinski ( 2017 ) suggest a “multimodel inference” approach, which assesses the hypothesized relationship in a wide range of models.…”