Metaphor detection for improving representation in linguistic rules

doi:10.11772/j.issn.1001-9081.2024071037

Abstract

Abstract:

The existing research work on metaphor detection task mostly adopts deep learning techniques and does not utilize linguistic rules in depth， which is mainly manifested in defective representation of semantic and basic meanings of target words involved in the rules， resulting in the related models’ inability to focus on differences between the target words and more relevant contextual words， and the boundaries between the basic meanings and the contextual meanings are still fuzzy. To address the above problems， a Metaphor Detection model to improve Representation in Linguistic rules （MeRL） was proposed. Firstly， semantic representation of the target words involved in both Selectional Preference Violation （SPV） and Metaphor Identification Procedure （MIP） rules was enhanced. Secondly， basic meanings of the target words in MIP rules were represented. Finally， the rule-based designed SPV and MIP modules were fused to identify metaphors jointly. Experimental results show that compared with MelBERT （Metaphor-aware late interaction over BERT） and other baseline models， the proposed model has the F1-score improved by at least 0.6， 0.9， and 1.2 percentage points， respectively， on benchmark datasets VUA-18， VUA Verb， and MOH-X， indicating that the proposed model achieves more accurate detection of metaphors. The proposed model has the F1-score improved by at least 0.7 percentage points when performing zero-shot transfer learning on TroFi dataset， indicating that the generalization ability of the proposed model is stronger.

Key words: metaphor, metaphor detection, pre-trained model, linguistic rule, external resource

摘要：

现有的隐喻识别任务研究工作多采用深度学习技术，而并未深入利用语言学规则，主要表现为规则中涉及的目标词的语义与基本义的表征存在缺陷，导致相关模型无法聚焦目标词与更相关上下文词之间的差异，且基本义与上下文含义界限仍然模糊。针对上述问题，提出一种改进语言规则中的表示的隐喻识别模型（MeRL）。首先，增强选择偏好违反（SPV）和隐喻识别过程（MIP）规则都涉及目标词的语义表示；其次，表征MIP规则中的目标词的基本义；最后，融合基于规则设计的SPV与MIP模块来共同识别隐喻。相较于MelBERT（Metaphor-aware late interaction over BERT）等基线模型，在基准数据集VUA-18、VUA Verb、MOH-X上的实验结果表明，所提模型的F1值分别至少提高了0.6、0.9、1.2个百分点，表明该模型识别隐喻更准确；在TroFi数据集上进行zero-shot迁移学习的结果显示，所提模型的F1值至少提高了0.7个百分点，表明该模型的泛化能力更强。

关键词: 隐喻, 隐喻识别, 预训练模型, 语言学规则, 外部资源

CLC Number:

TP391.1

Qing YANG, Yan ZHU. Metaphor detection for improving representation in linguistic rules[J]. Journal of Computer Applications, 2025, 45(8): 2491-2496.

杨青, 朱焱. 改进语言规则中的表示的隐喻识别[J]. 《计算机应用》唯一官方网站, 2025, 45(8): 2491-2496.

Figures/Tables 8

Fig. 1 Metaphor detection framework

Tab. 1 Detailed statistical data of benchmark datasets

数据集	目标词数	隐喻词比例%	句子数	句子的平均长度
VUA-18 train	116 622	11.2	6 323	18.4
VUA-18 val	38 628	11.6	1 550	24.9
VUA-18 test	50 175	12.4	2 694	18.6
VUA Verb train	15 516	27.9	7 479	20.2
VUA Verb val	1 724	26.9	1 541	25.0
VUA Verb test	2 694	30.0	2 694	18.6
MOH-X	647	48.7	647	8.0
TroFi	3 737	43.5	3 737	28.3

Tab. 2 Comparison of metaphor detection performance on three benchmark datasets

模型	VUA-18			VUA Verb			MOH-X（10 fold）
模型	P	R	F1	P	R	F1	P	R	F1
RNN_HG	74.5	73.1	73.8	67.9	71.3	69.5	78.3	80.6	79.2
RNN_MHCA	76.9	71.8	74.3	70.8	71.3	71.0	77.5	79.6	78.3
RoBERTa_SEQ	78.4	75.1	76.7	72.1	74.6	73.3	—	—	—
DeepMet	82.0	71.3	76.3	79.5	70.8	74.9	—	—	—
MelBERT	79.6	76.7	78.1	74.0	76.1	75.0	—	—	—
MrBERT	81.1	72.2	76.4	79.0	71.5	75.1	83.3	80.3	81.6
MisNet	82.0	73.5	77.5	72.6	76.8	74.6	88.5	76.7	82.1
MeRL	82.2	75.5	78.7	77.2	74.8	76.0	84.3	82.9	83.3

Tab. 3 Zero-shot transfer results on TroFi dataset

模型	P	R	F1
RoBERTa_SEQ	53.0	69.7	60.2
DeepMet	53.7	72.9	61.7
MelBERT	52.6	72.7	61.0
MrBERT	54.0	72.4	61.9
MisNet	53.9	73.2	62.1
MeRL	54.0	74.9	62.8

Tab. 4 Model performance of different types and parts of speech on VUA-18 dataset

类型	模型	P	R	F1	词性	模型	P	R	F1
学术	RNN_HG	80.0	78.5	79.2	动词	RNN_HG	67.9	72.0	69.9
	RNN_MHCA	84.2	77.4	80.6		RNN_MHCA	70.5	71.6	71.1
	RoBERTa_SEQ	84.3	80.6	82.4		RoBERTa_SEQ	72.3	74.9	73.6
	DeepMet	88.4	74.7	81.0		DeepMet	78.8	68.5	73.3
	MelBERT	86.3	80.3	83.2		MelBERT	74.3	76.5	75.4
	MisNet	87.6	78.2	82.6		MisNet	78.3	72.4	75.2
	MeRL	87.3	81.0	84.0		MeRL	78.2	74.0	76.1
对话	RNN_HG	65.1	67.5	66.3	形容词	RNN_HG	62.9	61.1	62.0
	RNN_MHCA	67.3	66.5	66.9		RNN_MHCA	69.9	53.2	60.4
	RoBERTa_SEQ	66.8	71.7	69.1		RoBERTa_SEQ	67.7	62.1	64.8
	DeepMet	71.6	71.1	71.4		DeepMet	79.0	52.9	63.3
	MelBERT	69.5	71.4	70.4		MelBERT	69.7	59.3	64.1
	MisNet	71.1	68.7	69.9		MisNet	73.2	57.5	64.4
	MeRL	71.6	69.6	70.6		MeRL	73.1	58.4	65.0
小说	RNN_HG	67.6	71.3	69.4	副词	RNN_HG	68.7	59.4	63.7
	RNN_MHCA	68.1	70.3	69.2		RNN_MHCA	79.8	56.9	66.5
	RoBERTa_SEQ	73.0	72.2	72.6		RoBERTa_SEQ	77.5	62.2	69.0
	DeepMet	76.1	70.1	73.0		DeepMet	79.4	66.4	72.3
	MelBERT	73.1	75.2	74.1		MelBERT	78.5	68.8	73.3
	MisNet	76.9	72.9	74.8		MisNet	80.8	63.9	71.4
	MeRL	75.3	71.5	73.3		MeRL	81.3	64.3	71.9
新闻	RNN_HG	77.2	70.3	73.6	名词	RNN_HG	69.6	58.0	63.3
	RNN_MHCA	78.9	68.5	73.3		RNN_MHCA	71.7	53.8	61.5
	RoBERTa_SEQ	81.4	71.6	76.2		RoBERTa_SEQ	74.0	62.6	67.8
	DeepMet	84.1	67.6	75.0		DeepMet	76.5	57.1	65.4
	MelBERT	81.0	75.7	78.2		MelBERT	75.8	63.8	69.3
	MisNet	84.3	70.5	76.8		MisNet	76.2	62.0	68.4
	MeRL	84.8	73.4	78.7		MeRL	77.1	63.1	69.4

Tab. 5 Ablation experimental results

模型	P	R	F1
MeRL	82.2	75.5	78.7
-MIP	80.9	73.6	77.0
-SPV	82.4	73.5	77.7
-Def	82.3	73.1	77.4
-Syn	74.5	80.2	77.3

Tab. 6 Results of different feature representation fusion methods

融合方法	P	R	F1
$(u, v)$	82.2	75.5	78.7
$u - v$	81.4	75.4	78.3
$(u * v)$	79.3	76.3	77.8

Tab. 6 Results of different feature representation fusion methods

融合方法	P	R	F1
$(u, v)$	82.2	75.5	78.7
$u - v$	81.4	75.4	78.3
$(u * v)$	79.3	76.3	77.8

Tab. 7 Error case analysis on VUA-18 dataset

真实标签	预测标签	句子
非隐喻	隐喻	1. *Design* ： Crossed lines over the toytown tram： City transport could soon be back on the right track， says Jonathan Glancey
非隐喻	隐喻	2.they’re treating alcohol as *food* .
非隐喻	隐喻	3.I had nothing particular planned， merely an idea that it might be interesting to thrash our *way* out into the open ocean， though， mindful of the danger of tropical storms， I had no intention of going too far from the safe shelter of a Bahamian hurricane hole.
隐喻	非隐喻	4.Right here， in this gaping mouth， lies the end of the *chain* .

References 25

[1]	WILKS Y. A preferential， pattern-seeking， semantics for natural language inference［J］. Artificial Intelligence， 1975， 6（1）： 53-74.
[2]	WILKS Y. Making preferences more active［J］. Artificial Intelligence， 1978， 11（3）： 197-223.
[3]	GROUP P. MIP： a method for identifying metaphorically used words in discourse［J］. Metaphor and Symbol， 2007， 22（1）： 1-39.
[4]	STEEN G J， DORST A G， HERRMANN J B， et al. A method for linguistic metaphor identification［M］. Amsterdam： John Benjamins Publishing Company， 2010.
[5]	DEVLIN J， CHANG M W， LEE K， et al. BERT： pre-training of deep bidirectional transformers for language understanding［C］// Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics： Human Language Technologies， Volume 1 （Long and Short Papers）. Stroudsburg： ACL， 2019： 4171-4186.
[6]	LIU Y， OTT M， GOYAL N， et al. RoBERTa： a robustly optimized BERT pretraining approach［EB/OL］. ［2024-09-20］..
[7]	GE M， MAO R， CAMBRIA E. A survey on computational metaphor processing techniques： from identification， interpretation， generation to application［J］. Artificial Intelligence Review， 2023， 56（S2）： 1829-1895.
[8]	MAO R， LIN C， GUERIN F. End-to-end sequential metaphor identification inspired by linguistic theories［C］// Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. Stroudsburg： ACL， 2019： 3888-3898.
[9]	GAO G， CHOI E， CHOI Y， et al. Neural metaphor detection in context［C］// Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. Stroudsburg： ACL， 2018： 607-613.
[10]	SU C， FUKUMOTO F， HUANG X， et al. DeepMet： a reading comprehension paradigm for token-level metaphor detection［C］// Proceedings of the 2nd Workshop on Figurative Language Processing. Stroudsburg： ACL， 2020： 30-39.
[11]	SONG W， ZHOU S， FU R， et al. Verb metaphor detection via contextual relation learning［C］// Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing （Volume 1： Long Papers）. Stroudsburg： ACL， 2021： 4240-4251.
[12]	LI Y， WANG S， LIN C， et al. Metaphor detection via explicit basic meanings modelling［C］// Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics （Volume 2： Short Papers）. Stroudsburg： ACL， 2023： 91-100.
[13]	CHOI M， LEE S， CHOI E， et al. MelBERT： metaphor detection via contextualized late interaction using metaphorical identification theories［C］// Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics： Human Language Technologies. Stroudsburg： ACL， 2021： 1763-1773.
[14]	ZHANG S， LIU Y. Metaphor detection via linguistics enhanced Siamese network［C］// Proceedings of the 29th International Conference on Computational Linguistics. ［S.l.］： International Committee on Computational Linguistics， 2022： 4149-4159.
[15]	WANG S， LI Y， LIN C， et al. Metaphor detection with effective context denoising［C］// Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics. Stroudsburg： ACL， 2023： 1404-1409.
[16]	SU C， WU K， CHEN Y. Enhanced metaphor detection via incorporation of external knowledge based on linguistic theories［C］// Findings of the Association for Computational Linguistics： ACL-IJCNLP 2021. Stroudsburg： ACL， 2021： 1280-1287.
[17]	RADFORD A， WU J， CHILD R， et al. Language models are unsupervised multitask learners［EB/OL］. ［2024-09-20］..
[18]	HONNIBAL M. spaCy 2： Natural language understanding with bloom embeddings， convolutional neural networks and incremental parsing［EB/OL］.［2024-09-20］. .
[19]	胡壮麟. 语言学教程（修订版中译本）［M］. 北京：北京大学出版社， 2002：69.
	HU Z L. Linguistics course （revised Chinese translation）［M］. Beijing： Peking University Press， 2002：69.
[20]	LEONG C W， KLEBANOV B B， SHUTOVA E. A report on the 2018 VUA metaphor detection shared task［C］// Proceedings of the 2018 Workshop on Figurative Language Processing. Stroudsburg： ACL， 2018： 56-66.
[21]	BIRKE J， SARKAR A. A clustering approach for nearly unsupervised recognition of nonliteral language［C］// Proceedings of the 11th Conference of the European Chapter of the Association for Computational Linguistics. Stroudsburg： ACL， 2006： 329-336.
[22]	BIRKE J， SARKAR A. Active learning for the identification of nonliteral language［C］// Proceedings of the 2007 Workshop on Computational Approaches to Figurative Language. Stroudsburg： ACL， 2007： 21-28.
[23]	MOHAMMAD S， SHUTOVA E， TURNEY P. Metaphor as a medium for emotion： an empirical study［C］// Proceedings of the 5th Joint Conference on Lexical and Computational Semantics. Stroudsburg： ACL， 2016： 23-33.
[24]	LEONG C W， KLEBANOV B B， HAMILL C， et al. A report on the 2020 VUA and TOEFL metaphor detection shared task［C］// Proceedings of the 2nd Workshop on Figurative Language Processing. Stroudsburg： ACL， 2020： 18-29.
[25]	REIMERS N， GUREVYCH I. Sentence-BERT： sentence embeddings using Siamese BERT-networks［C］// Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing. Stroudsburg： ACL， 2019： 3982-3992.