Search

Wikipedia's perceived high quality and broad language coverage have established it as a fundamental resource in multilingual NLP. In the context of low-resource languages, however, these quality assumptions are increasingly being scrutinised. This paper critically examines the data quality of Wikipedia in a non-English setting by subjecting it to various quality filtering techniques, revealing widespread issues such as a high percentage of one-line articles and duplicate articles. We evaluate the downstream impact of quality filtering on Wikipedia and find that data quality pruning is an effective means for resource-efficient training without hurting performance, especially for low-resource languages. Moreover, we advocate for a shift in perspective from seeking a general definition of data quality towards a more language- and task-specific one. Ultimately, we aim for this study to serve as a guide to using Wikipedia for pretraining in a multilingual setting.

Pixel-based language models have emerged as a compelling alternative to subword-based language modelling, particularly because they can represent virtually any script. PIXEL, a canonical example of such a model, is a vision transformer that has been pre-trained on rendered text. While PIXEL has shown promising cross-script transfer abilities and robustness to orthographic perturbations, it falls short of outperforming monolingual subword counterparts like BERT in most other contexts. This discrepancy raises questions about the amount of linguistic knowledge learnt by these models and whether their performance in language tasks stems more from their visual capabilities than their linguistic ones. To explore this, we probe PIXEL using a variety of linguistic and visual tasks to assess its position on the vision-to-language spectrum. Our findings reveal a substantial gap between the model's visual and linguistic understanding. The lower layers of PIXEL predominantly capture superficial visual features, whereas the higher layers gradually learn more syntactic and semantic abstractions. Additionally, we examine variants of PIXEL trained with different text rendering strategies, discovering that introducing certain orthographic constraints at the input level can facilitate earlier learning of surface-level features. With this study, we hope to provide insights that aid the further development of pixel-based language models., Comment: 9 pages, Accepted to EMNLP 2025 Main

POS tagging plays a fundamental role in numerous applications. While POS taggers are highly accurate in well-resourced settings, they lag behind in cases of limited or missing training data. This paper focuses on POS tagging for languages with limited data. We seek to identify the characteristics of datasets that make them favourable for training POS tagging models without using any labelled training data from the target language. This is a zero-shot approach. We compare the accuracies of a multilingual large language model (mBERT) fine-tuned on one or more languages related to the target language. Additionally, we compare these results with models trained directly on the target language itself. We do this for three target low-resource languages. Our research highlights the importance of accurate dataset selection for effective zero-shot POS tagging. Particularly, a strong linguistic relationship and high-quality datasets ensure optimal results. For extremely low-resource languages, zero-shot models prove to be a viable option., Comment: To appear at the 4th Multilingual NLP workshop collocated with EMNLP 2024

The development of monolingual language models for low and mid-resource languages continues to be hindered by the difficulty in sourcing high-quality training data. In this study, we present a novel cross-lingual vocabulary transfer strategy, trans-tokenization, designed to tackle this challenge and enable more efficient language adaptation. Our approach focuses on adapting a high-resource monolingual LLM to an unseen target language by initializing the token embeddings of the target language using a weighted average of semantically similar token embeddings from the source language. For this, we leverage a translation resource covering both the source and target languages. We validate our method with the Tweeties, a series of trans-tokenized LLMs, and demonstrate their competitive performance on various downstream tasks across a small but diverse set of languages. Additionally, we introduce Hydra LLMs, models with multiple swappable language modeling heads and embedding tables, which further extend the capabilities of our trans-tokenization strategy. By designing a Hydra LLM based on the multilingual model TowerInstruct, we developed a state-of-the-art machine translation model for Tatar, in a zero-shot manner, completely bypassing the need for high-quality parallel data. This breakthrough is particularly significant for low-resource languages like Tatar, where high-quality parallel data is hard to come by. By lowering the data and time requirements for training high-quality models, our trans-tokenization strategy allows for the development of LLMs for a wider range of languages, especially those with limited resources. We hope that our work will inspire further research and collaboration in the field of cross-lingual vocabulary transfer and contribute to the empowerment of languages on a global scale., Comment: Accepted at COLM 2024

Beyond individual languages, multilingual natural language processing (NLP) research increasingly aims to develop models that perform well across languages generally. However, evaluating these systems on all the world's languages is practically infeasible. To attain generalizability, representative language sampling is essential. Previous work argues that generalizable multilingual evaluation sets should contain languages with diverse typological properties. However, 'typologically diverse' language samples have been found to vary considerably in this regard, and popular sampling methods are flawed and inconsistent. We present a language sampling framework for selecting highly typologically diverse languages given a sampling frame, informed by language typology. We compare sampling methods with a range of metrics and find that our systematic methods consistently retrieve more typologically diverse language selections than previous methods in NLP. Moreover, we provide evidence that this affects generalizability in multilingual model evaluation, emphasizing the importance of diverse language sampling in NLP evaluation.

The NLP research community has devoted increased attention to languages beyond English, resulting in considerable improvements for multilingual NLP. However, these improvements only apply to a small subset of the world's languages. Aiming to extend this, an increasing number of papers aspires to enhance generalizable multilingual performance across languages. To this end, linguistic typology is commonly used to motivate language selection, on the basis that a broad typological sample ought to imply generalization across a broad range of languages. These selections are often described as being 'typologically diverse'. In this work, we systematically investigate NLP research that includes claims regarding 'typological diversity'. We find there are no set definitions or criteria for such claims. We introduce metrics to approximate the diversity of language selection along several axes and find that the results vary considerably across papers. Crucially, we show that skewed language selection can lead to overestimated multilingual performance. We recommend future work to include an operationalization of 'typological diversity' that empirically justifies the diversity of language samples., Comment: EMNLP 2024: Main Conference

Emotion classification is a challenging task in NLP due to the inherent idiosyncratic and subjective nature of linguistic expression, especially with code-mixed data. Pre-trained language models (PLMs) have achieved high performance for many tasks and languages, but it remains to be seen whether these models learn and are robust to the differences in emotional expression across languages. Sociolinguistic studies have shown that Hinglish speakers switch to Hindi when expressing negative emotions and to English when expressing positive emotions. To understand if language models can learn these associations, we study the effect of language on emotion prediction across 3 PLMs on a Hinglish emotion classification dataset. Using LIME and token level language ID, we find that models do learn these associations between language choice and emotional expression. Moreover, having code-mixed data present in the pre-training can augment that learning when task-specific data is scarce. We also conclude from the misclassifications that the models may overgeneralise this heuristic to other infrequent examples where this sociolinguistic phenomenon does not apply., Comment: 5 pages, Accepted to SIGTYP 2024 @ EACL

Creoles represent an under-explored and marginalized group of languages, with few available resources for NLP research.While the genealogical ties between Creoles and a number of highly-resourced languages imply a significant potential for transfer learning, this potential is hampered due to this lack of annotated data. In this work we present CreoleVal, a collection of benchmark datasets spanning 8 different NLP tasks, covering up to 28 Creole languages; it is an aggregate of novel development datasets for reading comprehension, relation classification, and machine translation for Creoles, in addition to a practical gateway to a handful of preexisting benchmarks. For each benchmark, we conduct baseline experiments in a zero-shot setting in order to further ascertain the capabilities and limitations of transfer learning for Creoles. Ultimately, we see CreoleVal as an opportunity to empower research on Creoles in NLP and computational linguistics, and in general, a step towards more equitable language technology around the globe., Comment: Accepted to TACL

Van Miltenburg et al. (2021) suggest NLP research should adopt preregistration to prevent fishing expeditions and to promote publication of negative results. At face value, this is a very reasonable suggestion, seemingly solving many methodological problems with NLP research. We discuss pros and cons -- some old, some new: a) Preregistration is challenged by the practice of retrieving hypotheses after the results are known; b) preregistration may bias NLP toward confirmatory research; c) preregistration must allow for reclassification of research as exploratory; d) preregistration may increase publication bias; e) preregistration may increase flag-planting; f) preregistration may increase p-hacking; and finally, g) preregistration may make us less risk tolerant. We cast our discussion as a dialogue, presenting both sides of the debate., Comment: EACL 2023

Language models are defined over a finite set of inputs, which creates a vocabulary bottleneck when we attempt to scale the number of supported languages. Tackling this bottleneck results in a trade-off between what can be represented in the embedding matrix and computational issues in the output layer. This paper introduces PIXEL, the Pixel-based Encoder of Language, which suffers from neither of these issues. PIXEL is a pretrained language model that renders text as images, making it possible to transfer representations across languages based on orthographic similarity or the co-activation of pixels. PIXEL is trained to reconstruct the pixels of masked patches instead of predicting a distribution over tokens. We pretrain the 86M parameter PIXEL model on the same English data as BERT and evaluate on syntactic and semantic tasks in typologically diverse languages, including various non-Latin scripts. We find that PIXEL substantially outperforms BERT on syntactic and semantic processing tasks on scripts that are not found in the pretraining data, but PIXEL is slightly weaker than BERT when working with Latin scripts. Furthermore, we find that PIXEL is more robust than BERT to orthographic attacks and linguistic code-switching, further confirming the benefits of modelling language with pixels., Comment: ICLR 2023

In recent years, the natural language processing (NLP) community has given increased attention to the disparity of efforts directed towards high-resource languages over low-resource ones. Efforts to remedy this delta often begin with translations of existing English datasets into other languages. However, this approach ignores that different language communities have different needs. We consider a group of low-resource languages, Creole languages. Creoles are both largely absent from the NLP literature, and also often ignored by society at large due to stigma, despite these languages having sizable and vibrant communities. We demonstrate, through conversations with Creole experts and surveys of Creole-speaking communities, how the things needed from language technology can change dramatically from one language to another, even when the languages are considered to be very similar to each other, as with Creoles. We discuss the prominent themes arising from these conversations, and ultimately demonstrate that useful language technology cannot be built without involving the relevant community., Comment: LREC 2022

Various efforts in the Natural Language Processing (NLP) community have been made to accommodate linguistic diversity and serve speakers of many different languages. However, it is important to acknowledge that speakers and the content they produce and require, vary not just by language, but also by culture. Although language and culture are tightly linked, there are important differences. Analogous to cross-lingual and multilingual NLP, cross-cultural and multicultural NLP considers these differences in order to better serve users of NLP systems. We propose a principled framework to frame these efforts, and survey existing and potential strategies., Comment: ACL 2022 - Theme track

In this work, we investigate the knowledge learned in the embeddings of multimodal-BERT models. More specifically, we probe their capabilities of storing the grammatical structure of linguistic data and the structure learned over objects in visual data. To reach that goal, we first make the inherent structure of language and visuals explicit by a dependency parse of the sentences that describe the image and by the dependencies between the object regions in the image, respectively. We call this explicit visual structure the \textit{scene tree}, that is based on the dependency tree of the language description. Extensive probing experiments show that the multimodal-BERT models do not encode these scene trees.Code available at \url{https://github.com/VSJMilewski/multimodal-probes}., Comment: Accepted at ACL 2022

Large multilingual pretrained language models such as mBERT and XLM-RoBERTa have been found to be surprisingly effective for cross-lingual transfer of syntactic parsing models (Wu and Dredze 2019), but only between related languages. However, source and training languages are rarely related, when parsing truly low-resource languages. To close this gap, we adopt a method from multi-task learning, which relies on automated curriculum learning, to dynamically optimize for parsing performance on outlier languages. We show that this approach is significantly better than uniform and size-proportional sampling in the zero-shot setting., Comment: ACL 2022

With an increase of dataset availability, the potential for learning from a variety of data sources has increased. One particular method to improve learning from multiple data sources is to embed the data source during training. This allows the model to learn generalizable features as well as distinguishing features between datasets. However, these dataset embeddings have mostly been used before contextualized transformer-based embeddings were introduced in the field of Natural Language Processing. In this work, we compare two methods to embed datasets in a transformer-based multilingual dependency parser, and perform an extensive evaluation. We show that: 1) embedding the dataset is still beneficial with these models 2) performance increases are highest when embedding the dataset at the encoder level 3) unsurprisingly, we confirm that performance increases are highest for small datasets and datasets with a low baseline score. 4) we show that training on the combination of all datasets performs similarly to designing smaller clusters based on language-relatedness., Comment: Accepted to TLT at SyntaxFest 2021

Creole languages such as Nigerian Pidgin English and Haitian Creole are under-resourced and largely ignored in the NLP literature. Creoles typically result from the fusion of a foreign language with multiple local languages, and what grammatical and lexical features are transferred to the creole is a complex process. While creoles are generally stable, the prominence of some features may be much stronger with certain demographics or in some linguistic situations. This paper makes several contributions: We collect existing corpora and release models for Haitian Creole, Nigerian Pidgin English, and Singaporean Colloquial English. We evaluate these models on intrinsic and extrinsic tasks. Motivated by the above literature, we compare standard language models with distributionally robust ones and find that, somewhat surprisingly, the standard language models are superior to the distributionally robust ones. We investigate whether this is an effect of over-parameterization or relative distributional stability, and find that the difference persists in the absence of over-parameterization, and that drift is limited, confirming the relative stability of creole languages., Comment: CoNLL 2021

This work introduces Itihasa, a large-scale translation dataset containing 93,000 pairs of Sanskrit shlokas and their English translations. The shlokas are extracted from two Indian epics viz., The Ramayana and The Mahabharata. We first describe the motivation behind the curation of such a dataset and follow up with empirical analysis to bring out its nuances. We then benchmark the performance of standard translation models on this corpus and show that even state-of-the-art transformer architectures perform poorly, emphasizing the complexity of the dataset., Comment: Fixed typo

Building robust natural language understanding systems will require a clear characterization of whether and how various linguistic meaning representations complement each other. To perform a systematic comparative analysis, we evaluate the mapping between meaning representations from different frameworks using two complementary methods: (i) a rule-based converter, and (ii) a supervised delexicalized parser that parses to one framework using only information from the other as features. We apply these methods to convert the STREUSLE corpus (with syntactic and lexical semantic annotations) to UCCA (a graph-structured full-sentence meaning representation). Both methods yield surprisingly accurate target representations, close to fully supervised UCCA parser quality---indicating that UCCA annotations are partially redundant with STREUSLE annotations. Despite this substantial convergence between frameworks, we find several important areas of divergence., Comment: COLING 2020 camera ready

We present K{\o}psala, the Copenhagen-Uppsala system for the Enhanced Universal Dependencies Shared Task at IWPT 2020. Our system is a pipeline consisting of off-the-shelf models for everything but enhanced graph parsing, and for the latter, a transition-based graph parser adapted from Che et al. (2019). We train a single enhanced parser model per language, using gold sentence splitting and tokenization for training, and rely only on tokenized surface forms and multilingual BERT for encoding. While a bug introduced just before submission resulted in a severe drop in precision, its post-submission fix would bring us to 4th place in the official ranking, according to average ELAS. Our parser demonstrates that a unified pipeline is effective for both Meaning Representation Parsing and Enhanced Universal Dependencies., Comment: IWPT shared task 2020

Transition-based and graph-based dependency parsers have previously been shown to have complementary strengths and weaknesses: transition-based parsers exploit rich structural features but suffer from error propagation, while graph-based parsers benefit from global optimization but have restricted feature scope. In this paper, we show that, even though some details of the picture have changed after the switch to neural networks and continuous representations, the basic trade-off between rich features and global optimization remains essentially the same. Moreover, we show that deep contextualized word embeddings, which allow parsers to pack information about global sentence structure into local feature representations, benefit transition-based parsers more than graph-based parsers, making the two approaches virtually equivalent in terms of both accuracy and error profile. We argue that the reason is that these representations help prevent search errors and thereby allow transition-based parsers to better exploit their inherent strength of making accurate local decisions. We support this explanation by an error analysis of parsing experiments on 13 languages., Comment: Accepted at EMNLP 2019

There is a growing interest in investigating what neural NLP models learn about language. A prominent open question is the question of whether or not it is necessary to model hierarchical structure. We present a linguistic investigation of a neural parser adding insights to this question. We look at transitivity and agreement information of auxiliary verb constructions (AVCs) in comparison to finite main verbs (FMVs). This comparison is motivated by theoretical work in dependency grammar and in particular the work of Tesni\`ere (1959) where AVCs and FMVs are both instances of a nucleus, the basic unit of syntax. An AVC is a dissociated nucleus, it consists of at least two words, and an FMV is its non-dissociated counterpart, consisting of exactly one word. We suggest that the representation of AVCs and FMVs should capture similar information. We use diagnostic classifiers to probe agreement and transitivity information in vectors learned by a transition-based neural parser in four typologically different languages. We find that the parser learns different information about AVCs and FMVs if only sequential models (BiLSTMs) are used in the architecture but similar information when a recursive layer is used. We find explanations for why this is the case by looking closely at how information is learned in the network and looking at what happens with different dependency representations of AVCs. We conclude that there may be benefits to using a recursive layer in dependency parsing and that we have not yet found the best way to integrate it in our parsers., Comment: Accepted by the Computational Linguistics journal

The need for tree structure modelling on top of sequence modelling is an open issue in neural dependency parsing. We investigate the impact of adding a tree layer on top of a sequential model by recursively composing subtree representations (composition) in a transition-based parser that uses features extracted by a BiLSTM. Composition seems superfluous with such a model, suggesting that BiLSTMs capture information about subtrees. We perform model ablations to tease out the conditions under which composition helps. When ablating the backward LSTM, performance drops and composition does not recover much of the gap. When ablating the forward LSTM, performance drops less dramatically and composition recovers a substantial part of the gap, indicating that a forward LSTM and composition capture similar information. We take the backward LSTM to be related to lookahead features and the forward LSTM to the rich history-based features both crucial for transition-based parsers. To capture history-based information, composition is better than a forward LSTM on its own, but it is even better to have a forward LSTM as part of a BiLSTM. We correlate results with language properties, showing that the improved lookahead of a backward LSTM is especially important for head-final languages., Comment: Accepted at NAACL 2019

We present the Uppsala system for the CoNLL 2018 Shared Task on universal dependency parsing. Our system is a pipeline consisting of three components: the first performs joint word and sentence segmentation; the second predicts part-of- speech tags and morphological features; the third predicts dependency trees from words and tags. Instead of training a single parsing model for each treebank, we trained models with multiple treebanks for one language or closely related languages, greatly reducing the number of models. On the official test run, we ranked 7th of 27 teams for the LAS and MLAS metrics. Our system obtained the best scores overall for word segmentation, universal POS tagging, and morphological features., Comment: Proceedings of the CoNLL 2018 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies

Punctuation is a strong indicator of syntactic structure, and parsers trained on text with punctuation often rely heavily on this signal. Punctuation is a diversion, however, since human language processing does not rely on punctuation to the same extent, and in informal texts, we therefore often leave out punctuation. We also use punctuation ungrammatically for emphatic or creative purposes, or simply by mistake. We show that (a) dependency parsers are sensitive to both absence of punctuation and to alternative uses; (b) neural parsers tend to be more sensitive than vintage parsers; (c) training neural parsers without punctuation outperforms all out-of-the-box parsers across all scenarios where punctuation departs from standard punctuation. Our main experiments are on synthetically corrupted data to study the effect of punctuation in isolation and avoid potential confounds, but we also show effects on out-of-domain data., Comment: Analyzing and interpreting neural networks for NLP, EMNLP 2018 workshop

We provide a comprehensive analysis of the interactions between pre-trained word embeddings, character models and POS tags in a transition-based dependency parser. While previous studies have shown POS information to be less important in the presence of character models, we show that in fact there are complex interactions between all three techniques. In isolation each produces large improvements over a baseline system using randomly initialised word embeddings only, but combining them quickly leads to diminishing returns. We categorise words by frequency, POS tag and language in order to systematically investigate how each of the techniques affects parsing quality. For many word categories, applying any two of the three techniques is almost as good as the full combined system. Character models tend to be more important for low-frequency open-class words, especially in morphologically rich languages, while POS tags can help disambiguate high-frequency function words. We also show that large character embedding sizes help even for languages with small character sets, especially in morphologically rich languages., Comment: EMNLP 2018

Previous work has suggested that parameter sharing between transition-based neural dependency parsers for related languages can lead to better performance, but there is no consensus on what parameters to share. We present an evaluation of 27 different parameter sharing strategies across 10 languages, representing five pairs of related languages, each pair from a different language family. We find that sharing transition classifier parameters always helps, whereas the usefulness of sharing word and/or character LSTM parameters varies. Based on this result, we propose an architecture where the transition classifier is shared, and the sharing of word and character parameters is controlled by a parameter that can be tuned on validation data. This model is linguistically motivated and obtains significant improvements over a monolingually trained baseline. We also find that sharing transition classifier parameters helps when training a parser on unrelated language pairs, but we find that, in the case of unrelated languages, sharing too many parameters does not help., Comment: EMNLP 2018

How to make the most of multiple heterogeneous treebanks when training a monolingual dependency parser is an open question. We start by investigating previously suggested, but little evaluated, strategies for exploiting multiple treebanks based on concatenating training sets, with or without fine-tuning. We go on to propose a new method based on treebank embeddings. We perform experiments for several languages and show that in many cases fine-tuning and treebank embeddings lead to substantial improvements over single treebanks or concatenation, with average gains of 2.0--3.5 LAS points. We argue that treebank embeddings should be preferred due to their conceptual simplicity, flexibility and extensibility., Comment: 7 pages. Accepted to ACL 2018, short papers

This thesis presents a study about the integration of information about Multiword Expressions (MWEs) into parsing with Combinatory Categorial Grammar (CCG). We build on previous work which has shown the benefit of adding information about MWEs to syntactic parsing by implementing a similar pipeline with CCG parsing. More specifically, we collapse MWEs to one token in training and test data in CCGbank, a corpus which contains sentences annotated with CCG derivations. Our collapsing algorithm however can only deal with MWEs when they form a constituent in the data which is one of the limitations of our approach. We study the effect of collapsing training and test data. A parsing effect can be obtained if collapsed data help the parser in its decisions and a training effect can be obtained if training on the collapsed data improves results. We also collapse the gold standard and show that our model significantly outperforms the baseline model on our gold standard, which indicates that there is a training effect. We show that the baseline model performs significantly better on our gold standard when the data are collapsed before parsing than when the data are collapsed after parsing which indicates that there is a parsing effect. We show that these results can lead to improved performance on the non-collapsed standard benchmark although we fail to show that it does so significantly. We conclude that despite the limited settings, there are noticeable improvements from using MWEs in parsing. We discuss ways in which the incorporation of MWEs into parsing can be improved and hypothesize that this will lead to more substantial results. We finally show that turning the MWE recognition part of the pipeline into an experimental part is a useful thing to do as we obtain different results with different recognizers., Comment: MSc thesis, The University of Edinburgh, 2014, School of Informatics, MSc Artificial Intelligence