COLING 2024 Papers

Evaluation datasets are critical resources for measuring the quality of pretrained language models. However, due to the high cost of dataset annotation, these resources are scarce for most languages other than English, making it difficult to assess the quality of language models. In this work, we present a new method for evaluation dataset construction which enables any language with a New Testament translation to receive a suite of evaluation datasets suitable for pretrained language model evaluation. The method critically involves aligning verses with those in the New Testament portion of English OntoNotes, and then projecting annotations from English to the target language, with no manual annotation required. We apply this method to 1051 New Testament translations in 859 languages and make them publicly available. Additionally, we conduct experiments which demonstrate the efficacy of our method for creating evaluation tasks which can assess language model quality.

End-to-end (E2E) approach is gradually replacing hybrid models for automatic speech recognition (ASR) tasks. However, the optimization of E2E models lacks an intuitive method for handling decoding shifts, especially in scenarios with a large number of domain-specific rare words that hold specific important meanings. Furthermore, the absence of knowledge-intensive speech datasets in academia has been a significant limiting factor, and the commonly used speech corpora exhibit significant disparities with realistic conversation. To address these challenges, we present Medical Interview (MED-IT), a multi-turn consultation speech dataset that contains a substantial number of knowledge-intensive named entities. We also explore methods to enhance the recognition performance of rare words for E2E models. We propose a novel approach, post-decoder biasing, which constructs a transform probability matrix based on the distribution of training transcriptions. This guides the model to prioritize recognizing words in the biasing list. In our experiments, for subsets of rare words appearing in the training speech between 10 and 20 times, and between 1 and 5 times, the proposed method achieves a relative improvement of 9.3% and 5.1%, respectively.

In this work, we introduce a pioneering research challenge: evaluating positive and potentially harmful messages within music products. We initiate by setting a multi-faceted, multi-task benchmark for music content assessment. Subsequently, we introduce an efficient multi-task predictive model fortified with ordinality-enforcement to address this challenge. Our findings reveal that the proposed method not only significantly outperforms robust task-specific alternatives but also possesses the capability to assess multiple aspects simultaneously. Furthermore, through detailed case studies, where we employed Large Language Models (LLMs) as surrogates for content assessment, we provide valuable insights to inform and guide future research on this topic. The code for dataset creation and model implementation is publicly available at https://github.com/RiTUAL-UH/music-message-assessment.

Populism is a phenomenon that is noticeably present in the political landscape of various countries over the past decades. While populism expressed by politicians has been thoroughly examined in the literature, populism expressed by citizens is still underresearched, especially when it comes to its automated detection in text. This work presents the PopAut corpus, which is the first annotated corpus of news comments for populism in the German language. It features 1,200 comments collected between 2019-2021 that are annotated for populist motives anti-elitism, people-centrism and people-sovereignty. Following the definition of Cas Mudde, populism is seen as a thin ideology. This work shows that annotators reach a high agreement when labeling news comments for these motives. The data set is collected to serve as the basis for automated populism detection using machine-learning methods. By using transformer-based models, we can outperform existing dictionaries tailored for automated populism detection in German social media content. Therefore our work provides a rich resource for future work on the classification of populist user comments in the German language.

Social media platforms are daily exhibiting millions of events. To preliminarily predict the mainstream public reaction to these events, we study trendy response prediction to automatically generate top-liked user replies to social media events. While previous works focus on generating responses without factoring in popularity, we propose Popularity-Aligned Language Models (PopALM) to distinguish responses liked by a larger audience through reinforcement learning. Recognizing the noisy labels from user “likes”, we tailor-make curriculum learning in proximal policy optimization (PPO) to help models capture the essential samples for easy-to-hard training. In experiments, we build a large-scale Weibo dataset for trendy response prediction, and its results show that PopALM can help boost the performance of advanced language models.

Polymers are widely used in diverse fields, and the demand for efficient methods to extract and organize information about them is increasing. An automated approach that utilizes machine learning can accurately extract relevant information from scientific papers, providing a promising solution for automating information extraction using annotated training data. In this paper, we introduce a polymer-relevant ontology featuring crucial entities and relations to enhance information extraction in the polymer science field. Our ontology is customizable to adapt to specific research needs. We present PolyNERE, a high-quality named entity recognition (NER) and relation extraction (RE) corpus comprising 750 polymer abstracts annotated using our ontology. Distinctive features of PolyNERE include multiple entity types, relation categories, support for various NER settings, and the ability to assert entities and relations at different levels. PolyNERE also facilitates reasoning in the RE task through supporting evidence. While our experiments with recent advanced methods achieved promising results, challenges persist in adapting NER and RE from abstracts to full-text paragraphs. This emphasizes the need for robust information extraction systems in the polymer domain, making our corpus a valuable benchmark for future developments.

In this paper, we present PolitiCAUSE, a new corpus of political texts annotated for causality. We provide a detailed and robust annotation scheme for annotating two types of information: (1) whether a sentence contains a causal relation or not, and (2) the spans of text that correspond to the cause and effect components of the causal relation. We also provide statistics and analysis of the corpus, and outline the difficulties and limitations of the task. Finally, we test out two transformer-based classification models on our dataset as a form of evaluation. The models achieve a moderate performance on the dataset, with a MCC score of 0.62. Our results show that PolitiCAUSE is a valuable resource for studying causality in texts, especially in the domain of political discourse, and that there is still room for improvement in developing more accurate and robust methods for this problem.

Aspect-Sentiment Triplet Extraction (ASTE) is one of the most challenging and complex tasks in sentiment analysis. It concerns the construction of triplets that contain an aspect, its associated sentiment polarity, and an opinion phrase that serves as a rationale for the assigned polarity. Despite the growing popularity of the task and the many machine learning methods being proposed to address it, the number of datasets for ASTE is very limited. In particular, no dataset is available for any of the Slavic languages. In this paper, we present two new datasets for ASTE containing customer opinions about hotels and purchased products expressed in Polish. We also perform experiments with two ASTE techniques combined with two large language models for Polish to investigate their performance and the difficulty of the assembled datasets. The new datasets are available under a permissive licence and have the same file format as the English datasets, facilitating their use in future research.

This paper presents the Polish Discourse Corpus, a pioneering resource of this kind for Polish and the first corpus in Poland to employ the ISO standard for discourse relation annotation. The Polish Discourse Corpus adopts ISO 24617-8, a segment of the Language Resource Management – Semantic Annotation Framework (SemAF), which outlines a set of core discourse relations adaptable for diverse languages and genres. The paper overviews the corpus architecture, annotation procedures, the challenges that the annotators have encountered, as well as key statistical data concerning discourse relations and connectives in the corpus. It further discusses the initial phases of the discourse parser tailored for the ISO 24617-8 framework. Evaluations on the efficacy and potential refinement areas of the corpus annotation and parsing strategies are also presented. The final part of the paper touches upon anticipated research plans to improve discourse analysis techniques in the project and to conduct discourse studies involving multiple languages.

Recently proposed systems for open-domain question answering (OpenQA) require large amounts of training data to achieve state-of-the-art performance. However, data annotation is known to be time-consuming and therefore expensive to acquire. As a result, the appropriate datasets are available only for a handful of languages (mainly English and Chinese). In this work, we introduce and publicly release PolQA, the first Polish dataset for OpenQA. It consists of 7,000 questions, 87,525 manually labeled evidence passages, and a corpus of over 7,097,322 candidate passages. Each question is classified according to its formulation, type, as well as entity type of the answer. This resource allows us to evaluate the impact of different annotation choices on the performance of the QA system and propose an efficient annotation strategy that increases the passage retrieval accuracy@10 by 10.55 p.p. while reducing the annotation cost by 82%.

In recent years, large language models have achieved state-of-the-art performance across various NLP tasks. However, investigations have shown that these models tend to rely on shortcut features, leading to inaccurate predictions and causing the models to be unreliable at generalization to out-of-distribution (OOD) samples. For instance, in the context of relation extraction (RE), we would expect a model to identify the same relation independently of the entities involved in it. For example, consider the sentence “Leonardo da Vinci painted the Mona Lisa” expressing the created(Leonardo_da_Vinci, Mona_Lisa) relation. If we substiute “Leonardo da Vinci” with “Barack Obama”, then the sentence still expresses the created relation. A robust model is supposed to detect the same relation in both cases. In this work, we describe several semantically-motivated strategies to generate adversarial examples by replacing entity mentions and investigate how state-of-the-art RE models perform under pressure. Our analyses show that the performance of these models significantly deteriorates on the modified datasets (avg. of -48.5% in F1), which indicates that these models rely to a great extent on shortcuts, such as surface forms (or patterns therein) of entities, without making full use of the information present in the sentences.

Although neural machine translation (NMT) models perform well in the general domain, it remains rather challenging to control their generation behavior to satisfy the requirement of different users. Given the expensive training cost and the data scarcity challenge of learning a new model from scratch for each user requirement, we propose a memory-augmented adapter to steer pretrained NMT models in a pluggable manner. Specifically, we construct a multi-granular memory based on the user-provided text samples and propose a new adapter architecture to combine the model representations and the retrieved results. We also propose a training strategy using memory dropout to reduce spurious dependencies between the NMT model and the memory. We validate our approach on both style- and domain-specific experiments and the results indicate that our method can outperform several representative pluggable baselines.

Generating visualizations from natural language queries is a useful extension to visualization libraries such as Vega-Lite. The goal of the NL2VIS task is to generate a valid Vega-Lite specification from a data frame and a natural language query as input, which can then be rendered as a visualization. To enable real-time interaction with the data, small model sizes and fast inferences are required. Previous work has introduced custom neural network solutions with custom visualization specifications and has not systematically tested pre-trained LMs to solve this problem. In this work, we opt for a more generic approach that (i) evaluates pre-trained LMs of different sizes and (ii) uses string encodings of data frames and visualization specifications instead of custom specifications. In our experiments, we show that these representations, in combination with pre-trained LMs, scale better than current state-of-the-art models. In addition, the small and base versions of the T5 architecture achieve real-time interaction, while LLMs far exceed latency thresholds suitable for visual exploration tasks. In summary, our models generate visualization specifications in real-time on a CPU and establish a new state of the art on the NL2VIS benchmark nvBench.

Received wisdom in linguistic typology holds that if the structure of a language becomes more complex in one dimension, it will simplify in another, building on the assumption that all languages are equally complex (Joseph and Newmeyer, 2012). We study this claim on a micro-level, using a tightly-controlled sample of Dutch dialects (across 366 collection sites) and Min dialects (across 60 sites), which enables a more fair comparison across varieties. Even at the dialect level, we find empirical evidence for a tradeoff between word length and a computational measure of phonotactic complexity from a LSTM-based phone-level language model—a result previously documented only at the language level. A generalized additive model (GAM) shows that dialects with low phonotactic complexity concentrate around the capital regions, which we hypothesize to correspond to prior hypotheses that language varieties of greater or more diverse populations show reduced phonotactic complexity. We also experiment with incorporating the auxiliary task of predicting syllable constituency, but do not find an increase in the strength of the negative correlation observed.

Research in speech technologies and comparative linguistics depends on access to diverse and accessible speech data. The UCLA Phonetics Lab Archive is one of the earliest multilingual speech corpora, with long-form audio recordings and phonetic transcriptions for 314 languages (Ladefoged et al., 2009). Recently, 95 of these languages were time-aligned with word-level phonetic transcriptions (Li et al., 2021). Here we present VoxAngeles, a corpus of audited phonetic transcriptions and phone-level alignments of the UCLA Phonetics Lab Archive, which uses the 95-language CMU re-release as our starting point. VoxAngeles also includes word- and phone-level segmentations from the original UCLA corpus, as well as phonetic measurements of word and phone durations, vowel formants, and vowel f0. This corpus enhances the usability of the original data, particularly for quantitative phonetic typology, as demonstrated through a case study of vowel intrinsic f0. We also discuss the utility of the VoxAngeles corpus for general research and pedagogy in crosslinguistic phonetics, as well as for low-resource and multilingual speech technologies. VoxAngeles is free to download and use under a CC-BY-NC 4.0 license.

Modeling interlocutor information is essential towards modeling multi-party conversations to account for the presence of multiple participants. We investigate the role of including the persona attributes of both the speaker and addressee relevant to each utterance, collected via 3 distinct mock social media experiments. The participants were recruited via MTurk, and were unaware of the persona attributes of the other users they interacted with on the platform. Our main contributions include 1) a multi-party conversation dataset with rich associated metadata (including persona), and 2) a persona-aware heterogeneous graph transformer response generation model. We find that PersonaHeterMPC provides a good baseline towards persona-aware generation for multi-party conversation modeling, generating responses which are relevant and consistent with the interlocutor personas relevant to the conversation.

Misogyny is often expressed through figurative language. Some neutral words can assume a negative connotation when functioning as pejorative epithets. Disambiguating the meaning of such terms might help the detection of misogyny. In order to address such task, we present PejorativITy, a novel corpus of 1,200 manually annotated Italian tweets for pejorative language at the word level and misogyny at the sentence level. We evaluate the impact of injecting information about disambiguated words into a model targeting misogyny detection. In particular, we explore two different approaches for injection: concatenation of pejorative information and substitution of ambiguous words with univocal terms. Our experimental results, both on our corpus and on two popular benchmarks on Italian tweets, show that both approaches lead to a major classification improvement, indicating that word sense disambiguation is a promising preliminary step for misogyny detection. Furthermore, we investigate LLMs’ understanding of pejorative epithets by means of contextual word embeddings analysis and prompting.

Identifying the type of relationship between words (cognates, borrowings, inherited) provides a deeper insight into the history of a language and allows for a better characterization of language relatedness. In this paper, we propose a computational approach for discriminating between cognates and borrowings, one of the most difficult tasks in historical linguistics. We compare the discriminative power of graphic and phonetic features and we analyze the underlying linguistic factors that prove relevant in the classification task. We perform experiments for pairs of languages in the Romance language family (French, Italian, Spanish, Portuguese, and Romanian), based on a comprehensive database of Romance cognates and borrowings. To our knowledge, this is one of the first attempts of this kind and the most comprehensive in terms of covered languages.

This paper presents a first attempt to apply Universal Dependencies (De Marneffe et al., 2021) to train a parser for Mauritian Creole (MC), a French-based Creole language spoken on the island of Mauritius. This paper demonstrates the construction of a 161-sentence (1007-token) treebank for MC and evaluates the performance of a part-of-speech tagger and Universal Dependencies parser trained on this data. The sentences were collected from publicly available grammar books (Syea, 2013) and online resources (Baker and Kriegel, 2013), as well as from government-produced school textbooks (Antonio-Françoise et al., 2021; Natchoo et al., 2017). The parser, trained with UDPipe 2 (Straka, 2018), reached F1 scores of UPOS=86.2, UAS=80.8 and LAS=69.8. This fares favorably when compared to models of similar size for other under-resourced Indigenous and Creole languages. We then address some of the challenges faced when applying UD to Creole languages in general and to Mauritian Creole in particular. The main challenge was the handling of spelling variation in the input. Other issues include the tagging of modal verbs, middle voice sentences, and parts of the tense-aspect-mood system (such as the particle fek).