English, like many languages, uses a wide variety of ways to talk about the future, which makes the automatic identification of future reference a challenge. In this research we ex- tend Latent Dirichlet allocation (LDA) for use in the identification of future-referring sentences. Building off a set of hand-designed rules, we trained a ADAGRAD classifier to be able to automatically detect sentences referring to the future. Uni-bi-trigram and syntactic rule mixed feature was found to provide the highest accuracy. Latent Dirichlet Allocation (LDA) indicated the existence of four major categories of future orientation. Lastly, the results of these analyses were found to correlate with a range of behavioral measures, offering evidence in support of the psychological reality of the categories.
Proceedings of the Annual Meeting of the Association for Computational Linguistics (ACL) / 2015
Anthology | Paper | Presentation | BibTeX