Aline Villavicencio

Maggie Mi, Golzar Atefi, Atsuki Yamaguchi et al.

Idioms can be analysed in terms of their decomposability, the extent to which constituent meanings contribute to the figurative whole. Decomposability is thought to predict syntactic flexibility. Usage-based accounts instead attribute idiom behaviour to distributional experience, such as speaker familiarity and predictability. We examine these views using contextualised language models as controlled distributional learners. We propose a model-internal measure of decomposability and relate it to human ratings, syntactic flexibility, and predictability while tracking idiom learning during pretraining. Model-derived decomposability correlates weakly with human judgments and shows a small but consistent negative relationship with syntactic flexibility. Pretraining analyses show that stabilisation of idiom representations in models is not explained by frequency alone. Instead, surprisal, decomposability, and frequency all contribute, with decomposability showing the strongest training-dependent effect.

Irina Bigoulaeva, Rachneet Sachdeva, Harish Tayyar Madabushi et al.

We compare sequential fine-tuning with a model for multi-task learning in the context where we are interested in boosting performance on two tasks, one of which depends on the other. We test these models on the FigLang2022 shared task which requires participants to predict language inference labels on figurative language along with corresponding textual explanations of the inference predictions. Our results show that while sequential multi-task learning can be tuned to be good at the first of two target tasks, it performs less well on the second and additionally struggles with overfitting. Our findings show that simple sequential fine-tuning of text-to-text models is an extraordinarily powerful method for cross-task knowledge transfer while simultaneously predicting multiple interdependent targets. So much so, that our best model achieved the (tied) highest score on the task.

Harish Tayyar Madabushi, Edward Gow-Smith, Marcos Garcia et al.

This paper presents the shared task on Multilingual Idiomaticity Detection and Sentence Embedding, which consists of two subtasks: (a) a binary classification task aimed at identifying whether a sentence contains an idiomatic expression, and (b) a task based on semantic text similarity which requires the model to adequately represent potentially idiomatic expressions in context. Each subtask includes different settings regarding the amount of training data. Besides the task description, this paper introduces the datasets in English, Portuguese, and Galician and their annotation procedure, the evaluation metrics, and a summary of the participant systems and their results. The task had close to 100 registered participants organised into twenty five teams making over 650 and 150 submissions in the practice and evaluation phases respectively.

Qinkai Yu, Chong Zhang, Gaojie Jin et al.

Annotating medical data for training AI models is often costly and limited due to the shortage of specialists with relevant clinical expertise. This challenge is further compounded by privacy and ethical concerns associated with sensitive patient information. As a result, well-trained medical segmentation models on private datasets constitute valuable intellectual property requiring robust protection mechanisms. Existing model protection techniques primarily focus on classification and generative tasks, while segmentation models-crucial to medical image analysis-remain largely underexplored. In this paper, we propose a novel, stealthy, and harmless method, StealthMark, for verifying the ownership of medical segmentation models under black-box conditions. Our approach subtly modulates model uncertainty without altering the final segmentation outputs, thereby preserving the model's performance. To enable ownership verification, we incorporate model-agnostic explanation methods, e.g. LIME, to extract feature attributions from the model outputs. Under specific triggering conditions, these explanations reveal a distinct and verifiable watermark. We further design the watermark as a QR code to facilitate robust and recognizable ownership claims. We conducted extensive experiments across four medical imaging datasets and five mainstream segmentation models. The results demonstrate the effectiveness, stealthiness, and harmlessness of our method on the original model's segmentation performance. For example, when applied to the SAM model, StealthMark consistently achieved ASR above 95% across various datasets while maintaining less than a 1% drop in Dice and AUC scores, significantly outperforming backdoor-based watermarking methods and highlighting its strong potential for practical deployment. Our implementation code is made available at: https://github.com/Qinkaiyu/StealthMark.

Edward Gow-Smith, Harish Tayyar Madabushi, Carolina Scarton et al.

Tokenisation is the first step in almost all NLP tasks, and state-of-the-art transformer-based language models all use subword tokenisation algorithms to process input text. Existing algorithms have problems, often producing tokenisations of limited linguistic validity, and representing equivalent strings differently depending on their position within a word. We hypothesise that these problems hinder the ability of transformer-based models to handle complex words, and suggest that these problems are a result of allowing tokens to include spaces. We thus experiment with an alternative tokenisation approach where spaces are always treated as individual tokens. Specifically, we apply this modification to the BPE and Unigram algorithms. We find that our modified algorithms lead to improved performance on downstream NLP tasks that involve handling complex words, whilst having no detrimental effect on performance in general natural language understanding tasks. Intrinsically, we find our modified algorithms give more morphologically correct tokenisations, in particular when handling prefixes. Given the results of our experiments, we advocate for always treating spaces as individual tokens as an improved tokenisation method.

Dilara Torunoğlu-Selamet, Dogukan Arslan, Rodrigo Wilkens et al.

Potentially idiomatic expressions (PIEs) construe meanings inherently tied to the everyday experience of a given language community. As such, they constitute an interesting challenge for assessing the linguistic (and to some extent cultural) capabilities of NLP systems. In this paper, we present XMPIE, a parallel multilingual and multimodal dataset of potentially idiomatic expressions. The dataset, containing 34 languages and over ten thousand items, allows comparative analyses of idiomatic patterns among language-specific realisations and preferences in order to gather insights about shared cultural aspects. This parallel dataset allows to evaluate model performance for a given PIE in different languages and whether idiomatic understanding in one language can be transferred to another. Moreover, the dataset supports the study of PIEs across textual and visual modalities, to measure to what extent PIE understanding in one modality transfers or implies in understanding in another modality (text vs. image). The data was created by language experts, with both textual and visual components crafted under multilingual guidelines, and each PIE is accompanied by five images representing a spectrum from idiomatic to literal meanings, including semantically related and random distractors. The result is a high-quality benchmark for evaluating multilingual and multimodal idiomatic language understanding.

Atsuki Yamaguchi, Terufumi Morishita, Aline Villavicencio et al.

Expanding the linguistic diversity of instruct large language models (LLMs) is crucial for global accessibility but is often hindered by the reliance on costly specialized target language labeled data and catastrophic forgetting during adaptation. We tackle this challenge under a realistic, low-resource constraint: adapting instruct LLMs using only unlabeled target language data. We introduce Source-Shielded Updates (SSU), a selective parameter update strategy that proactively preserves source knowledge. Using a small set of source data and a parameter importance scoring method, SSU identifies parameters critical to maintaining source abilities. It then applies a column-wise freezing strategy to protect these parameters before adaptation. Experiments across five typologically diverse languages and 7B and 13B models demonstrate that SSU successfully mitigates catastrophic forgetting. It reduces performance degradation on monolingual source tasks to just 3.4% (7B) and 2.8% (13B) on average, a stark contrast to the 20.3% and 22.3% from full fine-tuning. SSU also achieves target-language performance highly competitive with full fine-tuning, outperforming it on all benchmarks for 7B models and the majority for 13B models.

Dylan Phelps, Xuan-Rui Fan, Edward Gow-Smith et al.

Deep neural models, in particular Transformer-based pre-trained language models, require a significant amount of data to train. This need for data tends to lead to problems when dealing with idiomatic multiword expressions (MWEs), which are inherently less frequent in natural text. As such, this work explores sample efficient methods of idiomaticity detection. In particular we study the impact of Pattern Exploit Training (PET), a few-shot method of classification, and BERTRAM, an efficient method of creating contextual embeddings, on the task of idiomaticity detection. In addition, to further explore generalisability, we focus on the identification of MWEs not present in the training data. Our experiments show that while these methods improve performance on English, they are much less effective on Portuguese and Galician, leading to an overall performance about on par with vanilla mBERT. Regardless, we believe sample efficient methods for both identifying and representing potentially idiomatic MWEs are very encouraging and hold significant potential for future exploration.

Dylan Phelps, Rodrigo Wilkens, Edward Gow-Smith et al.

Alzheimer's Disease (AD) is a progressive neurodegenerative condition that adversely affects cognitive abilities. Language-related changes can be automatically identified through the analysis of outputs from linguistic assessment tasks, such as picture description. Language models show promise as a basis for screening tools for AD, but their limited interpretability poses a challenge in distinguishing true linguistic markers of cognitive decline from surface-level textual patterns. To address this issue, we examine how surface form variation affects classification performance, with the goal of assessing the ability of language models to represent underlying semantic indicators. We introduce a novel approach where texts surface forms are transformed by altering syntax and vocabulary while preserving semantic content. The transformations significantly modify the structure and lexical content, as indicated by low BLEU and chrF scores, yet retain the underlying semantics, as reflected in high semantic similarity scores, isolating the effect of semantic information, and finding models perform similarly to if they were using the original text, with only small deviations in macro-F1. We also investigate whether language from picture descriptions retains enough detail to reconstruct the original image using generative models. We found that image-based transformations add substantial noise reducing classification accuracy. Our methodology provides a novel way of looking at what features influence model predictions, and allows the removal of possible spurious correlations. We find that just using semantic information, language model based classifiers can still detect AD. This work shows that difficult to detect semantic impairment can be identified, addressing an overlooked feature of linguistic deterioration, and opening new pathways for early detection systems.

Marina Ribeiro, Bárbara Malcorra, Natália B. Mota et al.

Neurological disorders that affect speech production, such as Alzheimer's Disease (AD), significantly impact the lives of both patients and caregivers, whether through social, psycho-emotional effects or other aspects not yet fully understood. Recent advancements in Large Language Model (LLM) architectures have developed many tools to identify representative features of neurological disorders through spontaneous speech. However, LLMs typically lack interpretability, meaning they do not provide clear and specific reasons for their decisions. Therefore, there is a need for methods capable of identifying the representative features of neurological disorders in speech and explaining clearly why these features are relevant. This paper presents an explainable LLM method, named SLIME (Statistical and Linguistic Insights for Model Explanation), capable of identifying lexical components representative of AD and indicating which components are most important for the LLM's decision. In developing this method, we used an English-language dataset consisting of transcriptions from the Cookie Theft picture description task. The LLM Bidirectional Encoder Representations from Transformers (BERT) classified the textual descriptions as either AD or control groups. To identify representative lexical features and determine which are most relevant to the model's decision, we used a pipeline involving Integrated Gradients (IG), Linguistic Inquiry and Word Count (LIWC), and statistical analysis. Our method demonstrates that BERT leverages lexical components that reflect a reduction in social references in AD and identifies which further improve the LLM's accuracy. Thus, we provide an explainability tool that enhances confidence in applying LLMs to neurological clinical contexts, particularly in the study of neurodegeneration.

Maggie Mi, Aline Villavicencio, Nafise Sadat Moosavi

Human processing of idioms relies on understanding the contextual sentences in which idioms occur, as well as language-intrinsic features such as frequency and speaker-intrinsic factors like familiarity. While LLMs have shown high performance on idiomaticity detection tasks, this success may be attributed to reasoning shortcuts in existing datasets. To this end, we construct a novel, controlled contrastive dataset designed to test whether LLMs can effectively use context to disambiguate idiomatic meaning. Additionally, we explore how collocational frequency and sentence probability influence model performance. Our findings reveal that LLMs often fail to resolve idiomaticity when it is required to attend to the surrounding context, and that models perform better on sentences that have higher likelihood. The collocational frequency of expressions also impacts performance. We make our code and dataset publicly available.

Dylan Phelps, Rodrigo Wilkens, Edward Gow-Smith et al.

The recent trend towards utilisation of reasoning models has improved the performance of Large Language Models (LLMs) across many tasks which involve logical steps. One linguistic task that could benefit from this framing is idiomaticity detection, as a potentially idiomatic expression must first be understood before it can be disambiguated and serves as a basis for reasoning. In this paper, we explore how reasoning capabilities in LLMs affect idiomaticity detection performance and examine the effect of model size. We evaluate, as open source representative models, the suite of DeepSeek-R1 distillation models ranging from 1.5B to 70B parameters across four idiomaticity detection datasets. We find the effect of reasoning to be smaller and more varied than expected. For smaller models, producing chain-of-thought (CoT) reasoning increases performance from Math-tuned intermediate models, but not to the levels of the base models, whereas larger models (14B, 32B, and 70B) show modest improvements. Our in-depth analyses reveal that larger models demonstrate good understanding of idiomaticity, successfully producing accurate definitions of expressions, while smaller models often fail to output the actual meaning. For this reason, we also experiment with providing definitions in the prompts of smaller models, which we show can improve performance in some cases.

Atsuki Yamaguchi, Aline Villavicencio, Nikolaos Aletras

The development of state-of-the-art generative large language models (LLMs) disproportionately relies on English-centric tokenizers, vocabulary and pre-training data. Despite the fact that some LLMs have multilingual capabilities, recent studies have shown that their inference efficiency deteriorates when generating text in languages other than English. This results in increased inference time and costs. Cross-lingual vocabulary adaptation (CVA) methods have been proposed for adapting models to a target language aiming to improve downstream performance. However, the effectiveness of these methods on increasing inference efficiency of generative LLMs has yet to be explored. In this paper, we perform an empirical study of five CVA methods on four generative LLMs (including monolingual and multilingual models) across four typologically-diverse languages and four natural language understanding tasks. We find that CVA substantially contributes to LLM inference speedups of up to 271.5\%. We also show that adapting LLMs that have been pre-trained on more balanced multilingual data results in downstream performance comparable to the original models.

Dylan Phelps, Thomas Pickard, Maggie Mi et al.

Despite the recent ubiquity of large language models and their high zero-shot prompted performance across a wide range of tasks, it is still not known how well they perform on tasks which require processing of potentially idiomatic language. In particular, how well do such models perform in comparison to encoder-only models fine-tuned specifically for idiomaticity tasks? In this work, we attempt to answer this question by looking at the performance of a range of LLMs (both local and software-as-a-service models) on three idiomaticity datasets: SemEval 2022 Task 2a, FLUTE, and MAGPIE. Overall, we find that whilst these models do give competitive performance, they do not match the results of fine-tuned task-specific models, even at the largest scales (e.g. for GPT-4). Nevertheless, we do see consistent performance improvements across model scale. Additionally, we investigate prompting approaches to improve performance, and discuss the practicalities of using LLMs for these tasks.

Thomas Pickard, Aline Villavicencio, Maggie Mi et al.

Idiomatic expressions present a unique challenge in NLP, as their meanings are often not directly inferable from their constituent words. Despite recent advancements in Large Language Models (LLMs), idiomaticity remains a significant obstacle to robust semantic representation. We present datasets and tasks for SemEval-2025 Task 1: AdMiRe (Advancing Multimodal Idiomaticity Representation), which challenges the community to assess and improve models' ability to interpret idiomatic expressions in multimodal contexts and in multiple languages. Participants competed in two subtasks: ranking images based on their alignment with idiomatic or literal meanings, and predicting the next image in a sequence. The most effective methods achieved human-level performance by leveraging pretrained LLMs and vision-language models in mixture-of-experts settings, with multiple queries used to smooth over the weaknesses in these models' representations of idiomaticity.

Agne Knietaite, Adam Allsebrook, Anton Minkov et al.

Compositionality in language models presents a problem when processing idiomatic expressions, as their meaning often cannot be directly derived from their individual parts. Although fine-tuning and other optimization strategies can be used to improve representations of idiomatic expressions, this depends on the availability of relevant data. We present the Noun Compound Synonym Substitution in Books - NCSSB - datasets, which are created by substitution of synonyms of potentially idiomatic English noun compounds in public domain book texts. We explore the trade-off between data quantity and quality when training models for idiomaticity detection, in conjunction with contextual information obtained locally (from the surrounding sentences) or externally (through language resources). Performance on an idiomaticity detection task indicates that dataset quality is a stronger factor for context-enriched models, but that quantity also plays a role in models without context inclusion strategies.

Atsuki Yamaguchi, Terufumi Morishita, Aline Villavicencio et al.

Vocabulary expansion (VE) is the de-facto approach to language adaptation of large language models (LLMs) by adding new tokens and continuing pre-training on target data. While this is effective for base models trained on unlabeled data, it poses challenges for chat models trained to follow instructions through labeled conversation data. Directly adapting the latter with VE on target unlabeled data may result in forgetting chat abilities. While ideal, target chat data is often unavailable or costly to create for low-resource languages, and machine-translated alternatives are not always effective. To address this issue, previous work proposed using a base and chat model from the same family. This method first adapts the base LLM with VE on target unlabeled data and then converts it to a chat model by adding a chat vector (CV) derived from the weight difference between the source base and chat models. We propose ElChat, a new language adaptation method for chat LLMs that adapts a chat model directly on target unlabeled data, without a base model. It elicits chat abilities by injecting information from the source chat model. ElChat offers more robust and competitive target language and safety performance while achieving superior English, chat, and instruction-following abilities compared to CV.

Edward Gow-Smith, Dylan Phelps, Harish Tayyar Madabushi et al.

All existing transformer-based approaches to NLP using subword tokenisation algorithms encode whitespace (word boundary information) through the use of special space symbols (such as \#\# or \_) forming part of tokens. These symbols have been shown to a) lead to reduced morphological validity of tokenisations, and b) give substantial vocabulary redundancy. As such, removing these symbols has been shown to have a beneficial effect on the processing of morphologically complex words for transformer encoders in the pretrain-finetune paradigm. In this work, we explore whether word boundary information is at all useful to such models. In particular, we train transformer encoders across four different training scales, and investigate several alternative approaches to including word boundary information, evaluating on a range of tasks across different domains and problem set-ups: GLUE (for sentence-level classification), NER (for token-level classification), and two classification datasets involving complex words (Superbizarre and FLOTA). Overall, through an extensive experimental setup that includes the pre-training of 29 models, we find no substantial improvements from our alternative approaches, suggesting that modifying tokenisers to remove word boundary information isn't leading to a loss of useful information.

Wei He, Tiago Kramer Vieira, Marcos Garcia et al.

Idiomatic expressions are an integral part of human languages, often used to express complex ideas in compressed or conventional ways (e.g. eager beaver as a keen and enthusiastic person). However, their interpretations may not be straightforwardly linked to the meanings of their individual components in isolation and this may have an impact for compositional approaches. In this paper, we investigate to what extent word representation models are able to go beyond compositional word combinations and capture multiword expression idiomaticity and some of the expected properties related to idiomatic meanings. We focus on noun compounds of varying levels of idiomaticity in two languages (English and Portuguese), presenting a dataset of minimal pairs containing human idiomaticity judgments for each noun compound at both type and token levels, their paraphrases and their occurrences in naturalistic and sense-neutral contexts, totalling 32,200 sentences. We propose this set of minimal pairs for evaluating how well a model captures idiomatic meanings, and define a set of fine-grained metrics of Affinity and Scaled Similarity, to determine how sensitive the models are to perturbations that may lead to changes in idiomaticity. The results obtained with a variety of representative and widely used models indicate that, despite superficial indications to the contrary in the form of high similarities, idiomaticity is not yet accurately represented in current models. Moreover, the performance of models with different levels of contextualisation suggests that their ability to capture context is not yet able to go beyond more superficial lexical clues provided by the words and to actually incorporate the relevant semantic clues needed for idiomaticity.

Maggie Mi, Aline Villavicencio, Nafise Sadat Moosavi

Language models often struggle with idiomatic, figurative, or context-sensitive inputs, not because they produce flawed outputs, but because they misinterpret the input from the outset. We propose an input-only method for anticipating such failures using token-level likelihood features inspired by surprisal and the Uniform Information Density hypothesis. These features capture localized uncertainty in input comprehension and outperform standard baselines across five linguistically challenging datasets. We show that span-localized features improve error detection for larger models, while smaller models benefit from global patterns. Our method requires no access to outputs or hidden activations, offering a lightweight and generalizable approach to pre-generation error prediction.

Reyhaneh Hashempour, Barbara Plank, Aline Villavicencio et al.

This work presents a set of experiments conducted to predict the gender of Twitter users based on language-independent features extracted from the text of the users' tweets. The experiments were performed on a version of TwiSty dataset including tweets written by the users of six different languages: Portuguese, French, Dutch, English, German, and Italian. Logistic regression (LR), and feed-forward neural networks (FFNN) with back-propagation were used to build models in two different settings: Inter-Lingual (IL) and Cross-Lingual (CL). In the IL setting, the training and testing were performed on the same language whereas in the CL, Italian and German datasets were set aside and only used as test sets and the rest were combined to compose training and development sets. In the IL, the highest accuracy score belongs to LR whereas in the CL, FFNN with three hidden layers yields the highest score. The results show that neural network based models underperform traditional models when the size of the training set is small; however, they beat traditional models by a non-trivial margin, when they are fed with large enough data. Finally, the feature analysis confirms that men and women have different writing styles independent of their language.

Wei He, Marco Idiart, Carolina Scarton et al.

Accurately modeling idiomatic or non-compositional language has been a longstanding challenge in Natural Language Processing (NLP). This is partly because these expressions do not derive their meanings solely from their constituent words, but also due to the scarcity of relevant data resources, and their impact on the performance of downstream tasks such as machine translation and simplification. In this paper we propose an approach to model idiomaticity effectively using a triplet loss that incorporates the asymmetric contribution of components words to an idiomatic meaning for training language models by using adaptive contrastive learning and resampling miners to build an idiomatic-aware learning objective. Our proposed method is evaluated on a SemEval challenge and outperforms previous alternatives significantly in many metrics.

Atsuki Yamaguchi, Aline Villavicencio, Nikolaos Aletras

Large language models (LLMs) have shown remarkable capabilities in many languages beyond English. Yet, LLMs require more inference steps when generating non-English text due to their reliance on English-centric tokenizers and vocabulary, resulting in higher usage costs to non-English speakers. Vocabulary expansion with target language tokens is a widely used cross-lingual vocabulary adaptation approach to remedy this issue. Despite its effectiveness in inference speedup, previous work on vocabulary expansion has focused on high-resource settings assuming access to a substantial amount of target language data to effectively initialize the embeddings of the new tokens and adapt the LLM to the target language. However, vocabulary expansion in low-resource settings has yet to be explored. In this article, we investigate vocabulary expansion in low-resource settings by considering embedding initialization methods and continual pre-training strategies. Through extensive experiments across typologically diverse languages, tasks and models, we establish a set of strategies to perform vocabulary expansion for faster inference, while striving to maintain competitive downstream performance to baselines. This is achieved with only 30K sentences ($\sim$0.01GB text data) from the target language.

Kun Zhao, Bohao Yang, Chenghua Lin et al.

The long-standing one-to-many issue of the open-domain dialogues poses significant challenges for automatic evaluation methods, i.e., there may be multiple suitable responses which differ in semantics for a given conversational context. To tackle this challenge, we propose a novel learning-based automatic evaluation metric (CMN), which can robustly evaluate open-domain dialogues by augmenting Conditional Variational Autoencoders (CVAEs) with a Next Sentence Prediction (NSP) objective and employing Mutual Information (MI) to model the semantic similarity of text in the latent space. Experimental results on two open-domain dialogue datasets demonstrate the superiority of our method compared with a wide range of baselines, especially in handling responses which are distant to the golden reference responses in semantics.

Rodrigo Wilkens, Leonardo Zilio, Aline Villavicencio

Much recent effort has been devoted to creating large-scale language models. Nowadays, the most prominent approaches are based on deep neural networks, such as BERT. However, they lack transparency and interpretability, and are often seen as black boxes. This affects not only their applicability in downstream tasks but also the comparability of different architectures or even of the same model trained using different corpora or hyperparameters. In this paper, we propose a set of intrinsic evaluation tasks that inspect the linguistic information encoded in models developed for Brazilian Portuguese. These tasks are designed to evaluate how different language models generalise information related to grammatical structures and multiword expressions (MWEs), thus allowing for an assessment of whether the model has learned different linguistic phenomena. The dataset that was developed for these tasks is composed of a series of sentences with a single masked word and a cue phrase that helps in narrowing down the context. This dataset is divided into MWEs and grammatical structures, and the latter is subdivided into 6 tasks: impersonal verbs, subject agreement, verb agreement, nominal agreement, passive and connectors. The subset for MWEs was used to test BERTimbau Large, BERTimbau Base and mBERT. For the grammatical structures, we used only BERTimbau Large, because it yielded the best results in the MWE task.

Harish Tayyar Madabushi, Edward Gow-Smith, Carolina Scarton et al.

Despite their success in a variety of NLP tasks, pre-trained language models, due to their heavy reliance on compositionality, fail in effectively capturing the meanings of multiword expressions (MWEs), especially idioms. Therefore, datasets and methods to improve the representation of MWEs are urgently needed. Existing datasets are limited to providing the degree of idiomaticity of expressions along with the literal and, where applicable, (a single) non-literal interpretation of MWEs. This work presents a novel dataset of naturally occurring sentences containing MWEs manually classified into a fine-grained set of meanings, spanning both English and Portuguese. We use this dataset in two tasks designed to test i) a language model's ability to detect idiom usage, and ii) the effectiveness of a language model in generating representations of sentences containing idioms. Our experiments demonstrate that, on the task of detecting idiomatic usage, these models perform reasonably well in the one-shot and few-shot scenarios, but that there is significant scope for improvement in the zero-shot scenario. On the task of representing idiomaticity, we find that pre-training is not always effective, while fine-tuning could provide a sample efficient method of learning representations of sentences containing MWEs.

Marcely Zanon Boito, Bolaji Yusuf, Lucas Ondel et al.

Documenting languages helps to prevent the extinction of endangered dialects, many of which are otherwise expected to disappear by the end of the century. When documenting oral languages, unsupervised word segmentation (UWS) from speech is a useful, yet challenging, task. It consists in producing time-stamps for slicing utterances into smaller segments corresponding to words, being performed from phonetic transcriptions, or in the absence of these, from the output of unsupervised speech discretization models. These discretization models are trained using raw speech only, producing discrete speech units that can be applied for downstream (text-based) tasks. In this paper we compare five of these models: three Bayesian and two neural approaches, with regards to the exploitability of the produced units for UWS. For the UWS task, we experiment with two models, using as our target language the Mboshi (Bantu C25), an unwritten language from Congo-Brazzaville. Additionally, we report results for Finnish, Hungarian, Romanian and Russian in equally low-resource settings, using only 4 hours of speech. Our results suggest that neural models for speech discretization are difficult to exploit in our setting, and that it might be necessary to adapt them to limit sequence length. We obtain our best UWS results by using Bayesian models that produce high quality, yet compressed, discrete representations of the input speech signal.

Marcely Zanon Boito, Aline Villavicencio, Laurent Besacier

For endangered languages, data collection campaigns have to accommodate the challenge that many of them are from oral tradition, and producing transcriptions is costly. Therefore, it is fundamental to translate them into a widely spoken language to ensure interpretability of the recordings. In this paper we investigate how the choice of translation language affects the posterior documentation work and potential automatic approaches which will work on top of the produced bilingual corpus. For answering this question, we use the MaSS multilingual speech corpus (Boito et al., 2020) for creating 56 bilingual pairs that we apply to the task of low-resource unsupervised word segmentation and alignment. Our results highlight that the choice of language for translation influences the word segmentation performance, and that different lexicons are learned by using different aligned translations. Lastly, this paper proposes a hybrid approach for bilingual word segmentation, combining boundary clues extracted from a non-parametric Bayesian model (Goldwater et al., 2009a) with the attentional word segmentation neural model from Godard et al. (2018). Our results suggest that incorporating these clues into the neural models' input representation increases their translation and alignment quality, specially for challenging language pairs.

Marcely Zanon Boito, Aline Villavicencio, Laurent Besacier

For language documentation initiatives, transcription is an expensive resource: one minute of audio is estimated to take one hour and a half on average of a linguist's work (Austin and Sallabank, 2013). Recently, collecting aligned translations in well-resourced languages became a popular solution for ensuring posterior interpretability of the recordings (Adda et al. 2016). In this paper we investigate language-related impact in automatic approaches for computational language documentation. We translate the bilingual Mboshi-French parallel corpus (Godard et al. 2017) into four other languages, and we perform bilingual-rooted unsupervised word discovery. Our results hint towards an impact of the well-resourced language in the quality of the output. However, by combining the information learned by different bilingual models, we are only able to marginally increase the quality of the segmentation.

Alexandre Salle, Aline Villavicencio

In distributional semantics, the pointwise mutual information ($\mathit{PMI}$) weighting of the cooccurrence matrix performs far better than raw counts. There is, however, an issue with unobserved pair cooccurrences as $\mathit{PMI}$ goes to negative infinity. This problem is aggravated by unreliable statistics from finite corpora which lead to a large number of such pairs. A common practice is to clip negative $\mathit{PMI}$ ($\mathit{\texttt{-} PMI}$) at $0$, also known as Positive $\mathit{PMI}$ ($\mathit{PPMI}$). In this paper, we investigate alternative ways of dealing with $\mathit{\texttt{-} PMI}$ and, more importantly, study the role that negative information plays in the performance of a low-rank, weighted factorization of different $\mathit{PMI}$ matrices. Using various semantic and syntactic tasks as probes into models which use either negative or positive $\mathit{PMI}$ (or both), we find that most of the encoded semantics and syntax come from positive $\mathit{PMI}$, in contrast to $\mathit{\texttt{-} PMI}$ which contributes almost exclusively syntactic information. Our findings deepen our understanding of distributional semantics, while also introducing novel $PMI$ variants and grounding the popular $PPMI$ measure.

Marcely Zanon Boito, Aline Villavicencio, Laurent Besacier

Since Bahdanau et al. [1] first introduced attention for neural machine translation, most sequence-to-sequence models made use of attention mechanisms [2, 3, 4]. While they produce soft-alignment matrices that could be interpreted as alignment between target and source languages, we lack metrics to quantify their quality, being unclear which approach produces the best alignments. This paper presents an empirical evaluation of 3 main sequence-to-sequence models (CNN, RNN and Transformer-based) for word discovery from unsegmented phoneme sequences. This task consists in aligning word sequences in a source language with phoneme sequences in a target language, inferring from it word segmentation on the target side [5]. Evaluating word segmentation quality can be seen as an extrinsic evaluation of the soft-alignment matrices produced during training. Our experiments in a low-resource scenario on Mboshi and English languages (both aligned to French) show that RNNs surprisingly outperform CNNs and Transformer for this task. Our results are confirmed by an intrinsic evaluation of alignment quality through the use of Average Normalized Entropy (ANE). Lastly, we improve our best word discovery model by using an alignment entropy confidence measure that accumulates ANE over all the occurrences of a given alignment pair in the collection.

Marcely Zanon Boito, Antonios Anastasopoulos, Marika Lekakou et al.

This paper presents an extension to a very low-resource parallel corpus collected in an endangered language, Griko, making it useful for computational research. The corpus consists of 330 utterances (about 20 minutes of speech) which have been transcribed and translated in Italian, with annotations for word-level speech-to-transcription and speech-to-translation alignments. The corpus also includes morphosyntactic tags and word-level glosses. Applying an automatic unit discovery method, pseudo-phones were also generated. We detail how the corpus was collected, cleaned and processed, and we illustrate its use on zero-resource tasks by presenting some baseline results for the task of speech-to-translation alignment and unsupervised word discovery. The dataset is available online, aiming to encourage replicability and diversity in computational language documentation experiments.

Pierre Godard, Marcely Zanon-Boito, Lucas Ondel et al.

We present a first attempt to perform attentional word segmentation directly from the speech signal, with the final goal to automatically identify lexical units in a low-resource, unwritten language (UL). Our methodology assumes a pairing between recordings in the UL with translations in a well-resourced language. It uses Acoustic Unit Discovery (AUD) to convert speech into a sequence of pseudo-phones that is segmented using neural soft-alignments produced by a neural machine translation model. Evaluation uses an actual Bantu UL, Mboshi; comparisons to monolingual and bilingual baselines illustrate the potential of attentional word segmentation for language documentation.

Alexandre Salle, Aline Villavicencio

The positive effect of adding subword information to word embeddings has been demonstrated for predictive models. In this paper we investigate whether similar benefits can also be derived from incorporating subwords into counting models. We evaluate the impact of different types of subwords (n-grams and unsupervised morphemes), with results confirming the importance of subword information in learning representations of rare and out-of-vocabulary words.

Marcely Zanon Boito, Alexandre Berard, Aline Villavicencio et al.

Word discovery is the task of extracting words from unsegmented text. In this paper we examine to what extent neural networks can be applied to this task in a realistic unwritten language scenario, where only small corpora and limited annotations are available. We investigate two scenarios: one with no supervision and another with limited supervision with access to the most frequent words. Obtained results show that it is possible to retrieve at least 27% of the gold standard vocabulary by training an encoder-decoder neural machine translation system with only 5,157 sentences. This result is close to those obtained with a task-specific Bayesian nonparametric model. Moreover, our approach has the advantage of generating translation alignments, which could be used to create a bilingual lexicon. As a future perspective, this approach is also well suited to work directly from speech.

Alexandre Salle, Aline Villavicencio

Increasing the capacity of recurrent neural networks (RNN) usually involves augmenting the size of the hidden layer, with significant increase of computational cost. Recurrent neural tensor networks (RNTN) increase capacity using distinct hidden layer weights for each word, but with greater costs in memory usage. In this paper, we introduce restricted recurrent neural tensor networks (r-RNTN) which reserve distinct hidden layer weights for frequent vocabulary words while sharing a single set of weights for infrequent words. Perplexity evaluations show that for fixed hidden layer sizes, r-RNTNs improve language model performance over RNNs using only a small fraction of the parameters of unrestricted RNTNs. These results hold for r-RNTNs using Gated Recurrent Units and Long Short-Term Memory.

Alexandre Salle, Marco Idiart, Aline Villavicencio

In this paper we take a state-of-the-art model for distributed word representation that explicitly factorizes the positive pointwise mutual information (PPMI) matrix using window sampling and negative sampling and address two of its shortcomings. We improve syntactic performance by using positional contexts, and solve the need to store the PPMI matrix in memory by working on aggregate data in external memory. The effectiveness of both modifications is shown using word similarity and analogy tasks.

Alexandre Salle, Marco Idiart, Aline Villavicencio

In this paper, we propose LexVec, a new method for generating distributed word representations that uses low-rank, weighted factorization of the Positive Point-wise Mutual Information matrix via stochastic gradient descent, employing a weighting scheme that assigns heavier penalties for errors on frequent co-occurrences while still accounting for negative co-occurrence. Evaluation on word similarity and analogy tasks shows that LexVec matches and often outperforms state-of-the-art methods on many of these tasks.