Communicative pressures shape language during communication (not learning): Evidence from casemarking in artificial languages

Abstract

Natural languages are designed for efficient communication. A classic example is Differential Case Marking, when nouns are marked for their grammatical role only if this information cannot be derived from world knowledge (e.g. only atypical objects need to be linguistically marked as objects). Fedzechkina et al. (2012) present experimental evidence from an artificial language learning paradigm suggesting that biases in learning favour Differential Case Marking: learners exposed to a language with optional casemarking restructure the input, using casemarkers more in situations where marking would reduce ambiguity, despite the fact that they never use the artificial language in a communicative task where ambiguity matters. This is surprising given previous studies suggesting that biases in learning favour simplicity and are agnostic with respect to communicative function. We present 3 experiments investigating whether biases for communicatively-efficient Differential Case Marking exist in learning. Contrary to Fedzechkina et al. (2012), we find no evidence for such a bias in learning: participants’ use of case is impervious to the ambiguity of unmarked objects, and modulated by statistical properties of their input, observations which are inconsistent with their hypothesis and better explained by biases early in learning favouring iconicity in the form-meaning mapping, e.g. whereby atypical meanings are associated with atypical forms. However, we find good evidence that participants’ behaviour in actual communicative interaction are driven by efficient communication considerations: in interaction participants exhibited the expected Differential Object Marking pattern. This suggests that languages adapt to communicative efficiency constraints as a result of being used in communication, rather than due to biases in human learning favouring communicatively-efficient languages.

The impact of information structure on the emergence of differential object marking: an experimental study

“…The experiment was modeled after Fedzechkina et al (2012), using the materials and the design of Smith & Culbertson (2020), who replicated Fedzechkina et al (2012) online.…”

“…The procedure was adopted from Smith & Culbertson (2020). Participants were informed that they would learn an alien language called Smeespeak, taught by a monster named Smeeble.…”

“…2.1.5. Scoring Participants' word order (SOV or OSV) and use of object marking (present or absent) were automatically scored, adopting the procedure in Smith and Culbertson (2020). Each typed description was broken into words bounded by whitespace.…”

“…Moreover, learning from an ambiguous input (where there were no fixed roles for the animate characters) had no effect on participants' tendency to introduce DOM. This suggests that participants' apparent restructuring of the input language is not necessarily driven by a bias to efficiently reduce ambiguity (Smith & Culbertson, 2020) 2 . Instead, Smith & Culbertson (2020) found that participants' tendency to over-or undermark animate objects depended on the frequency of marking in the input, which is more consistent with a second explanation of DOM, which we call predictability-based marking accounts.…”

“…This suggests that participants' apparent restructuring of the input language is not necessarily driven by a bias to efficiently reduce ambiguity (Smith & Culbertson, 2020) 2 . Instead, Smith & Culbertson (2020) found that participants' tendency to over-or undermark animate objects depended on the frequency of marking in the input, which is more consistent with a second explanation of DOM, which we call predictability-based marking accounts. Finally, it is worth noting that the disambiguation explanation also fails to be clearly supported by diachronic evidence (Cristofaro, 2013(Cristofaro, , 2019.…”