To Share or not to Share: Predicting Sets of Sources for Model Transfer Learning

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 900–906 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 900 Voice Query Auto Completion Raphael Tang,1 Karun Kumar,1 Kendra Chalkley,∗2 Ji Xin,3 Liming Zhang,1 Wenyan Li,∗4 Gefei Yang,1 Yajie Mao,1 Junho Shin,1 G. Craig Murray,1 Jimmy Lin3 1Comcast AI 2SparkCognition Government Systems 3University of Waterloo 1firstname_lastname@comcast.com 2kendra.chalkley@sparkgov.ai 3{ji.xin,jimmylin}@uwaterloo.ca 4wenyanl62@gmail.com Abstract Query auto completion (QAC) is the task of predicting a search engine user’s final query from their intermediate, incomplete query. In this paper, we extend QAC to the streaming voice search setting, where automatic speech recognition systems produce intermediate tran- scripts as users speak. Naïvely applying ex- isting methods fails because the intermediate transcripts often don’t form prefixes or even substrings of the final transcript. To address this issue, we propose to condition QAC ap- proaches on intermediate transcripts to com- plete voice queries. We evaluate our models on a speech-enabled smart television with real- life voice search traffic, finding that this ASR- aware conditioning improves the completion quality. Our best method obtains an 18% rela- tive improvement in mean reciprocal rank over previous methods. 1 Introduction Query auto completion (QAC) is the task of pre- dicting a user’s complete query given the present, incomplete prefix of the query. For example, sup- pose a user types “COVID vaccine” into Google. Then, a QAC system proposes the most likely com- pletions for that prefix, e.g., “COVID vaccine near me,” saving the user time when the prediction is cor- rect. Existing state-of-the-art approaches generate completions using language models conditioned on the prefix (Park and Chiba, 2017), with simple prefix trees serving as a strong baseline. In the streaming voice search setting, such as on the Google Voice Assistant, naïvely adapting these existing approaches fails because the key assump- tions differ. Most glaringly, incomplete queries comprise partial speech, not text. In place of hu- man users, an automatic speech recognition (ASR) system produces the textual transcripts, resulting in intermediate queries that often don’t form pre- fixes or even substrings of the final query. Consider ∗ Work done while employed at Comcast AI. the voice query “Hulu” as an example. Passing it through a streaming ASR system, we observe the transcripts “who,” then “Hulu.” Traditional QAC approaches fail to complete “Hulu” from “who,” since they use orthographic prefixes and substrings (Cai and de Rijke, 2016) instead of the true phonetic prefix, which is generally unavailable at training time. Nevertheless, the intermediate transcript “who” is still informative toward predicting “Hulu” be- cause it frequently precedes “Hulu” in the sample. Based on this observation, we hypothesize that, for improved QAC quality, we must additionally model the dynamics between the intermediate and the final transcripts from the ASR system. In this paper, we precisely design and evaluate ASR-aware QAC models. The main contributions of our work are as follows: First, we are the first to describe the task of QAC for streaming, bidi- rectional ASR systems in voice search. Second, we propose and evaluate …

Content truncated to 500 words