Proceedings of the Annual Meeting of the Cognitive Science Society

As new concepts and discoveries accumulate over time, theamount of information available to speakers increases as well.One would expect that an utterance today would be more in-formative than an utterance 100 years ago (basing informationon surprisal; Shannon, 1948), given the increase in technol-ogy and scientific discoveries. This prediction, however, is atodds with recent theories regarding information in human lan-guage use, which suggest that speakers maintain a somewhatconstant information rate over time. Using the Google Ngramcorpus (Michel et al., 2011), we show for multiple languagesthat changes in lexical information (a unigram model) are actu-ally negatively correlated with changes in structural informa-tion (a trigram model), supporting recent proposals on infor-mation theoretic constraints.