Paper: Why Doesn't Natural Language Come Naturally?

ACL ID W99-0616
Title Why Doesn't Natural Language Come Naturally?
Venue 2000 Joint SIGDAT Conference on Empirical Methods in Natural Language Processing and Very Large Corpora
Session Main Conference
Year 1999

We have seen great success over the past 15 years in speech recognition. This success is due, largely, to the broad acceptance of Hidden Markov Models (HMMs) at the beginning of that period, which then facilitated rapid and steady improvements in speech recognition that are still continuing today. Although no one believes speech is produced by an HMM, the model affords a rich framework in which to improve the model in a rigorous and scientific manner. Could we create the same environment for a uni- form probabilistic paradigm in NL? It requires sev- eral ingredients: • A uniform notational system to express mean- ings, • A statistical model that can represent the asso- ciations between meanings and words, * A training program that estimates parameters from annotated examples, • An un...