2018/03/06 by Stefan Constantin, Constantin, Stefan, Jan Niehues +3
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Natural Language Processing Techniques #Speech and dialogue systems #Topic Modeling
paper · pdf · doi:10.48550/arxiv.1803.02279
openalex publication_date 2018/03/06 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Recently advancements in deep learning allowed the development of end-to-end\ntrained goal-oriented dialog systems. Although these systems already achieve\ngood performance, some simplifications limit their usage in real-life\nscenarios.\n In this work, we address two of these limitations: ignoring positional\ninformation and a fixed number of possible response candidates. We propose to\nuse positional encodings in the input to model the word order of the user\nutterances. Furthermore, by using a feedforward neural network, we are able to\ngenerate the output word by word and are no longer restricted to a fixed number\nof possible response candidates. Using the positional encoding, we were able to\nachieve better accuracies in the Dialog bAbI Tasks and using the feedforward\nneural network for generating the response, we were able to save computation\ntime and space consumption.\n