Speaker
Description
This talk picks up on work conducted with colleagues on the project 'Constraints on the Adaptiveness of Information in Language (CAIL)’,
which involved using information theory to analyze linguistic optionality and its cognitive scaffolding. Building on seminal work by Fenk and Fenk (1980, see also Fenk-Oczlon, 2001 and many subs), we suggest that linguistic planning is adapted for noise resistance. Specifically, speakers use whatever syntactic means are at their disposal in order to reduce the likelihood of catastrophic communication failure in the presence of noise. Thus, part of what motivates choice between syntactic alternatives is a type of risk mitigation.
First, we use real and permuted sentences from the Penn-York Computer-annotated Corpus of a Large Amount of English to demonstrate that more uniform ordering of elements confers functional noise resistance. Secondly, data from syntactic change in English and Icelandic (using the Penn Parsed Corpora of Historical English and the Icelandic Parsed Historical Corpus) shows that speakers use the syntactic variants made available by change in progress to approach a certain target or threshold of information uniformity, a threshold that is conserved over historical time. We have updated some prior work in this area on the OV-to-VO changes in English and Icelandic with additional work on the interaction between OV/VO, V2, and constraints on information spread.
Finally, I will present some data from the Penn-Helsinki Parsed Corpus of Early Modern English (Kroch et al 2004) and Penn Parsed Corpus of Modern British English (Kroch et al 2016) on the decline of DP topicalization in Late Early Modern English and its implications for information uniformity, carrying on the work of Speyer (2008, 2010). Surprisingly, object DP fronting appears to be on a trajectory of slow decline in modern English, quite independently of the well-known phrase structure changes in Middle and Early Modern English. I suggest that this is a "slow change" of the type described in Wallenberg (2016), and that fronted and in-situ orders are partially specialized along the continuous dimension of informational uniformity.
The existence of an information threshold or target is expected if the human language faculty constantly tries to keep the risk of information loss below a certain amount with a certain probability (not unlike the financial notion of Value at Risk), but at the same time, cannot achieve perfect uniformity due to linguistic constraints.
