Conversational Semantic Role Labeling with Predicate-Oriented Latent Graph Hao Fei1Shengqiong Wu1Meishan Zhang2Yafeng Ren3andDonghong Ji1 1Key Laboratory of Aerospace Information Security and Trusted Computing Ministry

2025-05-06 0 0 805.55KB 7 页 10玖币

侵权投诉

Conversational Semantic Role Labeling with Predicate-Oriented Latent Graph

Hao Fei1,Shengqiong Wu1,Meishan Zhang2,Yafeng Ren3∗and Donghong Ji1∗

1Key Laboratory of Aerospace Information Security and Trusted Computing, Ministry

of Education, School of Cyber Science and Engineering, Wuhan University, China

2Institute of Computing and Intelligence, Harbin Institute of Technology (Shenzhen), China

3Laboratory of Language and Artiﬁcial Intelligence, Guangdong University of Foreign Studies, China

{hao.fei, whuwsq, renyafeng, dhji}@whu.edu.cn, mason.zms@gmail.com

Abstract

Conversational semantic role labeling (CSRL) is

a newly proposed task that uncovers the shallow

semantic structures in a dialogue text. Unfor-

tunately several important characteristics of the

CSRL task have been overlooked by the exist-

ing works, such as the structural information in-

tegration, near-neighbor inﬂuence. In this work,

we investigate the integration of a latent graph

for CSRL. We propose to automatically induce

a predicate-oriented latent graph (POLar) with a

predicate-centered Gaussian mechanism, by which

the nearer and informative words to the predicate

will be allocated with more attention. The PO-

Lar structure is then dynamically pruned and re-

ﬁned so as to best ﬁt the task need. We addi-

tionally introduce an effective dialogue-level pre-

trained language model, CoDiaBERT, for better

supporting multiple utterance sentences and han-

dling the speaker coreference issue in CSRL. Our

system outperforms best-performing baselines on

three benchmark CSRL datasets with big margins,

especially achieving over 4% F1 score improve-

ments on the cross-utterance argument detection.

Further analyses are presented to better understand

the effectiveness of our proposed methods.

1 Introduction

Semantic Role Labeling (SRL) as a shallow semantic struc-

ture parsing task aims to ﬁnd all the arguments for a

given predicate [Gildea and Jurafsky, 2000; Marcheggiani

and Titov, 2017; Strubell et al., 2018; Fei et al., 2020d;

Fei et al., 2021b]. Conversational SRL (CSRL) is a newly

proposed task by Xu et al. [2021], which extends the reg-

ular SRL into multi-turn dialogue scenario. As illustrated

in Fig. 1, CSRL is characterized by that, the predicate is

given at current utterance, while the correlated arguments

are scattered in the history utterances of the dialogue that

are generated by two speakers. So far, few attempts have

been made for CSRL [Xu et al., 2021; Wu et al., 2021b;

Wu et al., 2021a], where, unfortunately, several key CSRL

∗Corresponding author

The leading role in The Age of Innocence ?

Just a supporting role, and via this movie won

an Oscar best female partner .

Foreign actresses really do will in acting for good works.

Winona Ryder , is really a beautiful woman.

Of course, because they are all running for the Oscar,

and how dare they not be serious.

ARG1

ARG3

ARG2

ARG0

Figure 1: Illustration of conversational SRL by two speakers. Word

‘won’ in yellow background is the predicate, linking to its different

types of arguments by arrows. The arugments in the same utterance

of the predicate are called intra-utterance arugment; those in differ-

ent dialogue turns are marked as cross-utterance arugment.

characteristics are still remained unexploted, which may

hamper the further task improvements.

First of all, intuitively SRL structure echoes much with

the syntactic dependency structure [Strubell et al., 2018;

Marcheggiani and Titov, 2017], and the existing regular SRL

works frequently employ external structural information for

performance enhancement, i.e., providing additional prior

links between predicates and arguments. However, it is quite

intractable to directly employ the external syntax knowledge

into CSRL for some reasons. For examples, a dependency

tree takes one single sentence piece as a unit, while a dia-

logue could contain multiple utterance sentences; the parse

trees from third-party parsers inevitably involve noises; only

a small part of the dependency structure can really offer helps,

rather than the entire tree [He et al., 2018].Second, the

predicate-argument structures in CSRL are broken down and

scattered into different utterances, which makes the detec-

tion of the CSRL more challenging. Actually the chances

are much higher for the predicate to ﬁnd its arguments when

they are being closer, i.e., near-neighbor inﬂuence. In other

words, nearer history utterances will show more impacts to

the latest utterance.1Fig. 1 exempliﬁes the case.

Based on the above observations, in this paper we present

an effective CSRL method with an innovative predicate-

1Our data statistics shows that, cross-one-utterance arguments

account for 60.3% among all cross-turn arguments; while the ratio

decreases to 30.3% for cross-two-utterance arguments.

arXiv:2210.03037v1 [cs.CL] 6 Oct 2022

oriented latent graph (namely, POLar). Unlike the explicit

syntactic structures, we make use of a two-parameter Hard-

Kuma distribution [Bastings et al., 2019]to automatically in-

duce latent graph from task’s need (cf. §4). Particularly, we

propose a predicate-centered Gaussian inducer for yielding

the latent edges, by which the nearer and informative words

to the predicate will be placed with more considerations. The

POLar is then dynamically pruned, so that only the task-

relevant structure will be built, while the irrelevant edges are

droped. The overall CSRL framework is differentiable and

performs predictions end-to-end (cf. Fig. 2).

The BERT [Devlin et al., 2019]pre-trained language

model (PLM) is extensively employed in existing works for

CSRL performance boosts [Xu et al., 2021; Wu et al., 2021a].

Nevertheless, it could be problematic to directly leverage

BERT for CSRL. On the one hand, one entire dialog often

consists of far more than two utterance sentences, while the

raw BERT restricts the input with at maximum two sentence

pieces, which consequently limits the PLM’s utility. There-

fore, we consider adopting the DiaBERT [Liu and Lapata,

2019; Li et al., 2020], which is designed for well supporting

multiple utterance inputs and thus yields better dialogue-level

representations. On the other hand, we note that in CSRL

both two speakers use the personal pronoun in their own per-

spective (i.e., ‘I’, ‘you’), and directly concatenating the multi-

turn utterances into PLM will unfortunately hurt the speaker-

role consistency, i.e., speaker coreference issue. Therefore,

we introduce a coreference-consistency-enhanced DiaBERT

(namely CoDiaBERT, cf. Fig. 3) that enhances the speaker-

role sensitivity of PLM with a pronoun-based speaker predic-

tion (PSP) strategy.

Our system signiﬁcantly outperforms strong-performing

baselines with big margins on three CSRL benchmarks. In

particular, over 4% F1 score of improvement is achieved for

detecting the cross-utterance type of arguments. Further anal-

yses reveal the usefulness of the proposed latent graph and the

dynamic pruning method, as well as the CoDiaBERT PLM.

Also we show that our model effectively solves long-range

dependence issue. Overall, we make these contributions:

•We for the ﬁrst time propose to improve the CSRL task

by incorporating a novel latent graph structure.

•We construct a predicate-oriented latent graph via a

predicate-centered Gaussian inducer. The structure is dynam-

ically pruned and reﬁned for best meeting the task need.

•We introduce a PLM for yielding better dialogue-level

text representations, which supports multiple utterance sen-

tences, and is sensitive to the speaker roles.

•Our framework achieves new state-of-the-art CSRL re-

sults on three benchmark data.

2 Related Work

The SRL task aims at uncovering the shallow semantic

structure of text, i.e. ‘who did what to whom where and

when’. As a fundamental natural language processing (NLP)

task, SRL can facilitate a broad range of downstream ap-

plications [Shen and Lapata, 2007; Liu and Gildea, 2010;

Wang et al., 2015]. By installing the current neural mod-

els, the current standard SRL has secured strong task per-

POLar

Induction

Dialogue

Encoder

CoDiaBERT

GCN x M

POLar

Pruning

Utterance1

<CLZ>

<SEP>

POLar

Encoder

word

…

<SEP>

…

prd

…

Utterance2

…

B-ARG0

I-ARG0

B-ARG1

Predicate node

Word node

Dialogue Decoder

Gating

⊕

Figure 2: The overall CSRL framework.

formances [Strubell et al., 2018; Li et al., 2019; Fei et

al., 2021c]. Recently, Xu et al. [2021]pioneer the task

of CSRL by extending the regular SRL into multi-turn di-

alogue scenario, in which they provide benchmark datasets

and CSRL neural model. Later a limited number of sub-

sequent works have explored this task [Wu et al., 2021b;

Wu et al., 2021a], where unfortunately several important fea-

tures of CSRL are not well considered. In this work, we im-

prove the CSRL by fully uncovering the task characteristics.

This work also closely relate to the line of syntax-driven

SRL [Marcheggiani and Titov, 2017; Fei et al., 2020c; Fei

et al., 2020b]. For the regular SRL, the external syntactic de-

pendency structure is a highly-frequently equipped feature for

performance enhancement, as the SRL shares much underly-

ing structure with syntax [He et al., 2018; Fei et al., 2020a;

Fei et al., 2021a]. However, it could be problematic for CSRL

to directly beneﬁt from such convient syntactic knowledge,

due to the dialogue nature of the text as we revealed ear-

lier. We thus propose to construct a latent structure at di-

alogue level, so as to facilitate the CSRL task with struc-

tural knowledge. In recent years, constructing latent graph for

downstream NLP tasks has received certain research attention

[Choi et al., 2018]. As an alternative to the pre-deﬁned syn-

tactic dependency structure yielded from third-party parsers,

latent structure induced from the task context could effec-

tively reduce noises [Corro and Titov, 2019], and meanwhile

enhance the efﬁcacy (i.e., creating task-relevant connections)

[Chen et al., 2020]. In this work, we revisit the characteristic

of CSRL, and based on the two-parameter Hard-Kuma distri-

bution [Bastings et al., 2019]investigate a predicate-oriented

latent graph by proposing a predicate-centered Gaussian in-

ducer.

3 CSRL Framework

Task modeling. Consider a conversation text U={ut}T

t=1

(Tis the total utterance number), with each utterance

ut={w0, w1,···} a sequence of words (w0is the utterance

speaker). In CSRL the predicate prd is labeled as input at the

current (lastest) utterance uT. We follow Xu et al. [2021],

modeling the task as a sequence labeling problem with a BIO

tagset. CSRL system identiﬁes and classiﬁes the arguments

of a predicate into semantic roles, such as A0,A1,AM-LOC,

etc, where we denote the complete role set as R. Given U

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

ConversationalSemanticRoleLabelingwithPredicate-OrientedLatentGraphHaoFei1,ShengqiongWu1,MeishanZhang2,YafengRen3andDonghongJi11KeyLaboratoryofAerospaceInformationSecurityandTrustedComputing,MinistryofEducation,SchoolofCyberScienceandEngineering,WuhanUniversity,China2InstituteofComputingandIntelli...

展开>> 收起<<

Conversational Semantic Role Labeling with Predicate-Oriented Latent Graph Hao Fei1Shengqiong Wu1Meishan Zhang2Yafeng Ren3andDonghong Ji1 1Key Laboratory of Aerospace Information Security and Trusted Computing Ministry.pdf

共7页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Conversational Semantic Role Labeling with Predicate-Oriented Latent Graph Hao Fei1Shengqiong Wu1Meishan Zhang2Yafeng Ren3andDonghong Ji1 1Key Laboratory of Aerospace Information Security and Trusted Computing Ministry

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: