Meta4XNLI: A Cross-lingual Parallel Corpus for Metaphor Detection and Interpretation

Metaphor is a conventional and ordinary part of language. A theory attempting to explain metaphor must account for the ease with which conventional metaphors are understood, and with the ability to understand novel metaphors as they are encountered. An approach to metaphor, based on the explicit representation of knowledge about metaphors, has been developed to address these issues. This approach asserts that the interpretation of conventional metaphoric language should proceed through the direct application of specific knowledge about the metaphors in the language. Correspondingly, the interpretation of novel metaphors can be accomplished through the systematic extension, elaboration, and combination of knowledge about already well-understood metaphors. MIDAS (Metaphor Interpretation, Denotation and Acquisition System) is a computer program that embodies this approach. MIDAS can be used to perform the following tasks: represent knowledge about conventional metaphors, interpret metaphoric language by applying this knowledge, and dynamically learn new metaphors as they are encountered during normal processing. Knowledge about conventional metaphors is represented in the form of coherent sets of associations between disparate conceptual domains. the representation captures both the details of individual metaphors, and the systematics exhibited by the set of metaphors in the language as a whole. These systematic sets of associations were implemented using the KODIAK knowledge representation language. MIDAS is capable of using this metaphoric knowledge to interpret conventional metaphoric language. The main thrust of this approach is that normal processing of metaphoric language proceeds through the direct application of specific knowledge about the metaphors in the language. This approach gives equal status to all conventional metaphoric and literal interpretations. Moreover, the mechanisms used to arrive at metaphoric and literal interpretations are fundamentally the same. When a metaphor is encountered for which MIDAS has no applicable knowledge, MIDAS calls upon its learning component - the Metaphor Extension System (MES). The approach embodies in the MES asserts that a novel metaphor can best be understood through the systematic extension of an already well-understood metaphor. MIDAS has been integrated as a part of the UNIX Consultant system. UC is a natural language consultant system that provides naive computer users with advice on how to use the UNIX operating system. By calling MIDAS, UC can successfully interpret and learn conventional UNIX domain metaphors, as they are encountered during the course of UC''s normal processing.

Powerful transformer models based on neural networks such as GPT‐4 have enabled huge progress in natural language processing. This paper identifies three challenges for computer programs dealing with metaphors. First, the phenomenon of Twice‐Apt‐Metaphors shows that metaphorical interpretations do not have to be triggered by syntactical, semantic or pragmatic tensions. The detection of these metaphors seems to involve a sense of aesthetic pleasure or a higher‐order theory of mind, both of which are difficult to implement into computer programs. Second, the contexts relative to which metaphors are interpreted are not simply given but must be reconstructed based on pragmatic considerations that can involve presuppositional pretence. If computer programs cannot produce or understand such a form of pretence, they will have problems dealing with certain metaphors. Finally, adequately interpreting and reacting to some metaphors seems to require the ability to have internal, first‐personal experiential and affective states. Since it is questionable whether computer programs have such mental states, it can be assumed that they will have problems with these kinds of metaphors.

The automatic detection of hate speech is a blooming field in the natural language processing community. In recent years there have been efforts in detecting hate speech in multiple languages, using models trained on multiple languages at the same time. Furthermore, there is special interest in the capabilities of language agnostic features to represent text in hate speech detection. This is because models can be trained in multiple languages, and then the capabilities of the model and representation can be tested on a unseen language.In this work we focused on detecting hate speech in mono-lingual, multi-lingual and cross-lingual settings. For this we used a pre-trained language model called Language Agnostic BERT Sentence Embeddings (LabSE), both for feature extraction and as an end to end classification model. We tested different models such as Support Vector Machines and Tree-based models, and representations in particular bag of words, bag of characters, and sentence embeddings extracted from Multi-lingual BERT. The dataset used was the SemEval 2019 task 5 data set, which covers hate speech against immigrants and women in English and Spanish. The results show that the usage of LabSE as feature extraction improves the performance on both languages in a mono-lingual setting, and in a cross-lingual setting. Moreover, LabSE as an end to end classification model performs better than the reported by the authors of SemEval 2019 task 5 data set for the Spanish language.

The rapid spread of rumors on social media and their potential impact has motivated the development of automatic rumor detection solutions. However, the existing solutions are mostly limited to detecting rumors in English which neglects the bulk of social media content in other low-resource languages. This paper aims to address the research gaps by proposing Multilingual Source Co-Attention Transformer (MUSCAT), which builds on a multilingual pre-trained language model to perform multilingual rumor detection. Specifically, MUSCAT pivots the source claims in multilingual conversation threads with co-attention transformers to improve detection performance in multilingual settings. We additionally construct multilingual rumor datasets to support our experimental evaluations. Our experimental results show that MUSCAT outperforms state-of-the-art methods in monolingual, cross-lingual, and multilingual rumor detection settings. We have also conducted empirical analysis and outlined the challenges of performing rumor detection in multilingual and cross-lingual settings.

Metaphor interpretation and use: A window into semantics in schizophrenia

ABSTRACTThe study examines factors underlying cross-language transfer in bilingual children; the main focus is on a measure of metaphor interpretation. Subjects were Spanish-English children ranging in age from 7 to 12 years. Measures were obtained for nonverbal mental capacity, metaphor interpretation, verbal-conceptual repertoire, and linguistic proficiency in English and Spanish. Using a previously validated procedure, subjects' metaphor interpretations were scored for cognitive complexity. In both languages, metaphor score was higher in older than in younger children. Correlational analyses indicated that level of metaphor interpretation was most strongly related to cognitive-developmental variables that are interdependent across languages, that is, nonverbal mental capacity and verbal-conceptual repertoire. Variables that measure specific proficiency in a language were less strongly related to level of metaphor interpretation, and did not exhibit cross-language correlations. This pattern was clearly seen in results of exploratory factor analyses. The role of cognitive versus linguistic factors in metaphor development is discussed, as is the issue of interdependence versus independence across first and second languages.

Metaphor is a figure of speech that describes one thing (a target) by mentioning another thing (a source) in a way that is not literally true. Metaphor understanding is an interesting but challenging problem in natural language processing. This paper presents a novel method for metaphor processing based on knowledge graph (KG) embedding. Conceptually, we abstract the structure of a metaphor as an attribute-dependent relation between the target and the source. Each specific metaphor can be represented as a metaphor triple (target, attribute, source). Therefore, we can model metaphor triples just like modeling fact triples in a KG and exploit KG embedding techniques to learn better representations of concepts, attributes and concept relations. In this way, metaphor interpretation and generation could be seen as KG completion, while metaphor detection could be viewed as a representation learning enhanced concept pair classification problem. Technically, we build a Chinese metaphor KG in the form of metaphor triples based on simile recognition, and also extract concept-attribute collocations to help describe concepts and measure concept relations. We extend the translation-based and the rotation-based KG embedding models to jointly optimize metaphor KG embedding and concept-attribute collocation embedding. Experimental results demonstrate the effectiveness of our method. Simile recognition is feasible for building the metaphor triple resource. The proposed models improve the performance on metaphor interpretation and generation, and the learned representations also benefit nominal metaphor detection compared with strong baselines.

Unlock the Potential of Counterfactually-Augmented Data in Out-Of-Distribution Generalization

Latent semantic similarity based interpretation of Chinese metaphors

Learning human languages is a difficult task for a computer. However, Deep Learning (DL) techniques have enhanced performance significantly for almost all-natural language processing (NLP) tasks. Unfortunately, these models cannot be generalized for all the NLP tasks with similar performance. NLU (Natural Language Understanding) is a subset of NLP including tasks, like machine translation, dialogue-based systems, natural language inference, text entailment, sentiment analysis, etc. The advancement in the field of NLU is the collective performance enhancement in all these tasks. Even though MTL (Multi-task Learning) was introduced before Deep Learning, it has gained significant attention in the past years. This paper aims to identify, investigate, and analyze various language models used in NLU and NLP to find directions for future research. The Systematic Literature Review (SLR) is prepared using the literature search guidelines proposed by Kitchenham and Charters on various language models between 2011 and 2021. This SLR points out that the unsupervised learning method-based language models show potential performance improvement. However, they face the challenge of designing the general-purpose framework for the language model, which will improve the performance of multi-task NLU and the generalized representation of knowledge. Combining these approaches may result in a more efficient and robust multi-task NLU. This SLR proposes building steps for a conceptual framework to achieve goals of enhancing the performance of language models in the field of NLU.

Counterfactually-Augmented Data (CAD) has the potential to improve language models’ Out-Of-Distribution (OOD) generalization capability, as CAD induces language models to exploit causal features and exclude spurious correlations. However, the empirical results of OOD generalization on CAD are not as efficient as expected. In this paper, we attribute the inefficiency to Myopia Phenomenon caused by CAD: language models only focus on causal features that are edited in the augmentation and exclude other non-edited causal features. As a result, the potential of CAD is not fully exploited. Based on the structural properties of CAD, we design two additional constraints to help language models extract more complete causal features contained in CAD, thus improving the OOD generalization capability. We evaluate our method on two tasks: Sentiment Analysis and Natural Language Inference, and the experimental results demonstrate that our method could unlock CAD’s potential and improve language models’ OOD generalization capability.

This paper analyzes the exchange between Ricoeur and Derrida concerning metaphor. I argue that the exchange is not a ＂missed encounter,＂ as Eftichis Pirovolakis has suggested, but exemplifies a hermeneutic situation in which theoretical divergence is supplemented by a practical convergence. Rather than a mere exegesis of the exchange between Ricoeur and Derrida, I emphasize the practical implications for the interpretation of poetic metaphors. To be more specific, I emphasize the case of Paul Celan's poem ＂Blume＂ and the semantic density of the central metaphor. Although Ricoeur and Derrida diverge in strictly theoretical terms, their theoretical positions-when translated into practical terms-establish different but convergent paradigms for the interpretation of poetic metaphors.

When girls are pearls, does it mean that they are beautiful or that they are pleasant? Not only are metaphors open to different interpretations but also these interpretations might vary across individuals, even with the same cultural context. However, the literature lacks a description of which patterns of interpretation emerge across individuals and which factors might drive them. Here, we investigated the role of multimodality, intended as the contribution of different dimensions of experience-based information, to explain individual variability in metaphor interpretation. We analyzed participants’ interpretations in a metaphor verbalization task according to a series of semantic features of words (affective, cognitive, and sensory) that mirror different cognitive mechanisms. With an innovative method that combines i) Natural Language Processing (NLP), ii) a multivariate statistical technique that derives Intersubject Representational Dissimilarity Matrixes (IS-RDMs), and iii) a data-driven clustering method, we were able to identify two groups of participants. One cluster, which we named mentalizers, exhibited greater use of cognitive and affective terms (e.g. the girls-pearls metaphors was explained as indicating that girls are pleasant), while the other cluster, which we named imagers, capitalized more on words expressing sensory-based features (e.g., girls were described as beautiful). Our study showed that a data-driven approach can capture different metaphor interpretative profiles from word-level semantic features and that differences are driven by the sensorimotor vs. sociocognitive dimensions. This suggest that there are multiple alternative routes to derive metaphorical meaning, involving different modality systems in the multimodal network for metaphor.

This research investigated interpretation of the second language metaphor in Persian-English bilinguals within the theory of conceptual metaphor founded by Lackoff and Johnson (1980). Because of the complex construction ground and cultural variance of conceptual metaphors, interpretation of some novel sentences or even ordinary language causes confusion to second language learners. Sixteen English metaphorical sentences based on four generic conceptual metaphors were selected and presented in a test paper with a forced-choice task and an explanatory task. Fourty Adults with different English proficiency level participated in the experiment. Results show that L2 proficiency poses effect on metaphor interpretation. Participants with higher English proficiency had better performance. Apart from English proficiency, different perceptions on ordinary experience and L1 and L2 lexicon size among individuals play a role in the metaphor interpretation process. Although the present study primarily aims at investigating personal properties taking part in the process, context in sentences is found to affect some subjects’ choices.

Transfer learning has become the dominant paradigm for many natural language processing tasks. In addition to models being pretrained on large datasets, they can be further trained on intermediate (supervised) tasks that are similar to the target task. For small Natural Language Inference (NLI) datasets, language modelling is typically followed by pretraining on a large (labelled) NLI dataset before fine-tuning with each NLI subtask. In this work, we explore Gradient Boosted Decision Trees (GBDTs) as an alternative to the commonly used Multi-Layer Perceptron (MLP) classification head. GBDTs have desirable properties such as good performance on dense, numerical features and are effective where the ratio of the number of samples w.r.t the number of features is low. In addition, we introduce FreeGBDT, a method of fitting a GBDT head on the features computed during fine-tuning to increase performance without additional computation by the neural network. We demonstrate the effectiveness of our method on several NLI datasets using a strong baseline model (RoBERTa-large with MNLI pretraining). The FreeGBDT shows a consistent improvement over the standard MLP classification head.