Frequency data from corpora partially explain native-speaker ratings and choices in overabundant paradigm cells

Abstract

Abstract:If we can operationalize corpus frequency in multiple ways, using absolute values and proportional values, which of them is more closely connected with the behaviour of language users? In this contribution, we examine overabundant cells in morphological paradigms, and look at the contribution that frequency of occurrence can make to understanding the choices speakers make due to this richness. We look at ways of operationalizing the term frequency in data from corpora and native speakers: the proportional frequency of forms (i.e. percentage of time that a variant is found in corpus data considered as a proportion of all variants) and several interpretations of absolute frequency (i.e. the raw frequency of variants in data from the same corpus). Working with data from unmotivated morphological variation in Czech case forms, we show that different instantiations of frequency help interpret the way variation is perceived and maintained by native speakers. Proportional frequency seems most salient for speakers in forming their judgements, while certain types of absolute frequency seem to have a dominant role in production tasks.Key words: corpus linguistics, frequency, morphology, empirical research, surveys, questionnaires, Czech, overabundance IntroductionFrequency data are familiar territory for any linguist who works with corpora. We cite the number of times a feature appears, or its normalized frequency if we are comparing corpora; we cite percentages to show structure within categories or to demonstrate change over time. Hidden behind the way we deal with these data is an implicit operationalization of our questions about language. We have chosen to let the corpus stand in for a particular language, type of language, genre, etc., but at the same time we have also chosen representations of frequency that give us the best chance of answering our research questions. It is worth interrogating these differing operationalizations of frequency to see how the same data, approached in different ways, can shed a different light on the way native speakers apprehend and use language.The term frequency is elastic, and once we start looking at frequency data there are few limits to the number of ways we can treat it. Divjak (2016) considers, among other meanings, the traditional relative frequency (incidence per million), construction frequency (which itself covers various ways of relating the frequencies between related items), family frequency (incorporating various ways of looking at the size and composition of a class of words) and measures of probability and association. These will largely be beyond the scope of this study, which is focused on how we understand and manipulate the numbers that arise from simple counts of individual forms.Our material comes from three sources. We have data from the Czech National Corpus (CNC) on the frequencies of forms occupying a single morphological "slot". We selected items

Do Users’ Reading Skills and Difficulty Ratings for Texts Affect Choices and Evaluations?

“…A skewed ratio of forms (say, 99:1) is treated differently than a more balanced ratio of forms (say, 5:1 or 3:1), and this operates regardless of which specific variant is in question. These findings are entirely in line with our previous investigations (Bermel & Knittl 2012a, 2012b, Bermel, Knittl & Russell 2015, 2017.…”

“…to test various hypotheses about frequency. In particular, Bermel, Knittl & Russell 2017 demonstrated that proportional frequency of forms had a consistent effect, at least on the sort of tasks we were asking respondents to perform.…”

“…For a number of years now, we have been looking at places in the Czech conjugational and declensional systems where a syntactic 'slot' has multiple exponents whose usage is not clearly differentiated, a situation described variously as competition (Lečić 2015), variation (Bermel & Knittl 2012a, 2012b, Bermel, Knittl & Russell 2015a, 2017 or overabundance (Thornton 2012).3…”

Schemas and the frequency/acceptability mismatch: Corpus distribution predicts sentence judgments

“…It must be kept in mind that schema compatibility is a conceptual notion that may be very difficult to determine for structural units with higher type frequencies and/or multi-facetted interactions with smaller units, especially in morphological contexts (Arppe and Järvikivi 2007;Bermel et al 2018;Divjak 2017). The identification of schema compatibility for go/come-V is straightforward, as the type frequency of two means that go/come-V is low in schematicity.…”

Intralingual Variation in Acceptability Judgments and Production: Three Case Studies in Russian Grammar

“…Another attempt to relate acceptability ratings and elicited usage data was by Bermel et al (2018), who retrieved probabilities from a balanced corpus and compared them to the distribution of existing options in fill-in-the-gap and rating tasks, completed by respondents simultaneously. Bermel et al (2018) took the responses to the fill-in-the-gap task to serve as production data; however, they observed that this could more accurately be thought of as a forced-choice situation, as there were only two possible options in the two syntactic contexts. Although a correlation was found, there is a limitation to this study, namely, the performance of two distinct tasks within a single questionnaire.…”