Prosodic segmentation for parsing spoken dialogue
8801Authors
nan
Abstract
AbstractParsing spoken dialogue poses unique difficulties, including disfluencies and unmarked boundaries between sentence-like units. Previous work has shown that prosody can help with parsing disfluent speech (Tran et al. 2018), but has assumed that the input to the parser is already segmented into sentence-like units (SUs), which isn’t true in existing speech applications. We investigate how prosody affects a parser that receives an entire dialogue turn as input (a turn-based model), instead of gold standard pre-segmented SUs (an SU-based model). In experiments on the English Switchboard corpus, we find that when using transcripts alone, the turn-based model has trouble segmenting SUs, leading to worse parse performance than the SU-based model. However, prosody can effectively replace gold standard SU boundaries: with prosody, the turn-based model performs as well as the SU-based model (91.38 vs. 91.06 F1 score, respectively), despite performing two tasks (SU segmentation and parsing) rather than one (parsing alone). Analysis shows that pitch and intensity features are the most important for this corpus, since they allow the model to correctly distinguish an SU boundary from a speech disfluency – a distinction that the model otherwise struggles to make.
Keywords
Full Content
Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 82–92 August 1–6, 2021. ©2021 Association for Computational Linguistics 82 PENS: A Dataset and Generic Framework for Personalized News Headline Generation Xiang Ao♠♦∗, Xiting Wang♥, Ling Luo♠♦, Ying Qiao♣, Qing He♠♦, Xing Xie♥† ♠Key Lab of Intelligent Information Processing of Chinese Academy of Sciences (CAS), Institute of Computing Technology, CAS, Beijing 100190, China ♥Microsoft Research Asia ♣Microsoft ♦University of Chinese Academy of Sciences, Beijing 100049, China {aoxiang,luoling18s,heqing}@ict.ac.cn {xitwan,yiqia,xing.xie}@microsoft.com Abstract In this paper, we formulate the personalized news headline generation problem whose goal is to output a user-specific title based on both a user’s reading interests and a candidate news body to be exposed to her. To build up a benchmark for this problem, we publicize a large-scale dataset named PENS (PErsonal- ized News headlineS). The training set is col- lected from user impressions logs of Microsoft News, and the test set is manually created by hundreds of native speakers to enable a fair testbed for evaluating models in an offline mode. We propose a generic framework as a preparatory solution to our problem. At its heart, user preference is learned by leveraging the user behavioral data, and three kinds of user preference injections are proposed to per- sonalize a text generator and establish person- alized headlines. We investigate our dataset by implementing several state-of-the-art user modeling methods in our framework to demon- strate a benchmark score for the proposed dataset. The dataset is available at https: //msnews.github.io/pens.html. 1 Introduction News headline generation (Dorr et al., 2003; Lopy- rev, 2015; Alfonseca et al., 2013; Tan et al., 2017; See et al., 2017; Zhang et al., 2018; Xu et al., 2019; Murao et al., 2019; Gavrilov et al., 2019; Gu et al., 2020; Song et al., 2020), conventionally considered as a paradigm of challenging text summarization task, has been extensively explored for decades. Their intuitive intention is to empower the model to output a condensed generalization, e.g., one sen- tence, of a news article. The recent year escalation of online content vendors such as Google News, TopBuzz, and ∗ This work was done when Xiang was visiting MSRA supported by the MSRA Young Visiting Researcher Program. †Corresponding author. etc (LaRocque, 2003) propels a new research di- rection that how to decorate the headline as an irre- sistible invitation to users for reading through the article (Xu et al., 2019) since more readings may ac- quaint more revenue of these platforms. To this end, specified stylized headline generation techniques were proposed, such as question headline (Zhang et al., 2018), sensational headline (Xu et al., 2019) generation, and so on (Shu et al., 2018; Gu et al., 2020). However, the over-decorate headlines might bring negative effects as click-baits begin to be- come notorious in ubiquitous online services1. Hence, the question is now changing to how to construct a title that catches on reader curios- ity without entering into click-bait territory. In- …
Content truncated to 500 words
Labeled Text
AbstractParsing spoken dialogue poses unique difficulties, including disfluencies and unmarked boundaries between sentence-like units. Previous work has shown that prosody can help with parsing disfluent speech (Tran et al. 2018), but has assumed that the input to the parser is already segmented into sentence-like units (SUs), which isn’t true in existing speech applications. We investigate how prosody affects a parser that receives an entire dialogue turn as input (a turn-based model), instead of gold standard pre-segmented SUs (an SU-based model). In experiments on the English Switchboard corpus, we find that when using transcripts alone, the turn-based model has trouble segmenting SUs, leading to worse parse performance than the SU-based model. However, prosody can effectively replace gold standard SU boundaries: with prosody, the turn-based model performs as well as the SU-based model (91.38 vs. 91.06 F1 score, respectively), despite performing two tasks (SU segmentation and parsing) rather than one (parsing alone). Analysis shows that pitch and intensity features are the most important for this corpus, since they allow the model to correctly distinguish an SU boundary from a speech disfluency – a distinction that the model otherwise struggles to make.
Publication Details
- Paper ID:
- 8801
- Year:
- 2021
- Publication:
- TIPSTER
- Publication Type:
- conf
- DOI:
- nan