Human Language in the Internet of Things

Important results from the project

Experience in text analysis at Gavagai has motivated more in depth study of how attitude is represented in text, how it is related to situational factors, and how it relates to referentiality and aspectuality in text. This present project has implemented and published an analysis model which enables a processing pipeline for such studies currently under execution. Technology from Gavagai has during the project period been presented for numerous potential future partners: research institutions, and companies both large and small.

Expected long term effects

The two major project outcomes so fare are (1) The project has implemented a model for including a richer linguistic feature palette in continuous machine learning processing models. This model has been presented in a number of international venues and results from studies both completed and under way will demonstrate how a richer feature palette enables faster learning curves and robust transfer between domains. (2) Gavagai has attained a much broader visibility and Gavagais main product is now applied in several research projects at Stanford.

Approach and implementation

The main activity of the project has been the design, implementation, and testing of an analysis model for human language data which allows the inclusion of linguistic analyses in a processing pipeline for continuous machine learning based on neural processing models. This work has proceeded largely according to original plan. Several studies for application to both new and established language technology tasks are under execution and some have already been reported in academic publications. Currently the model is being tested for large scale application in an industrial project.