Many Natural Language Processing (NLP) tasks are labeled on the token level, forthese tasks, the first step is to identify the tokens (tokenization). Becausethis step is often considered to be a solved problem, gold tokenization iscommonly assumed. In this paper, we propose an efficient method fortokenization with subword-based language models, and reflect on the status ofperformance on the tokenization task by evaluating on 122 languages in 20different scripts. We show that our proposed model performs on par with thestate-of-the-art, and that tokenization performance is mainly dependent on theamount and consistency of annotated data. We conclude that besidesinconsistencies in the data and exceptional cases the task can be consideredsolved for Latin languages for in-dataset settings (>99.5 F1). However,performance is 0.75 F1 point lower on average for datasets in other scripts andperformance deteriorates in cross-dataset setups.

In Natural Language Processing, entity linking (EL) has centered around Wikipedia, but yet remains underexplored for the job market domain. Disambiguating skill mentions can help us get insight into the current labor market demands. In this work, we are the first to explore EL in this domain, specifically targeting the linkage of occupational skills to the ESCO taxonomy (le Vrang et al., 2014). Previous efforts linked coarse-grained (full) sentences to a corresponding ESCO skill. In this work, we link more fine-grained span-level mentions of skills. We tune two high-performing neural EL models, a bi-encoder (Wu et al., 2020) and an autoregressive model (Cao et al., 2021), on a synthetically generated mention–skill pair dataset and evaluate them on a human-annotated skill-linking benchmark. Our findings reveal that both models are capable of linking implicit mentions of skills to their correct taxonomy counterparts. Empirically, BLINK outperforms GENRE in strict evaluation, but GENRE performs better in loose evaluation (accuracy@k).

Annotation tools are the starting point for creating Natural Language Processing (NLP) datasets. There is a wide variety of tools available; setting up these tools is however a hindrance. We propose Eevee, an annotation tool focused on simplicity, efficiency, and ease of use. It can run directly in the browser (no setup required) and uses tab-separated files (as opposed to character offsets or task-specific formats) for annotation. It allows for annotation of multiple tasks on a single dataset and supports four task-types: sequence labeling, span labeling, text classification and seq2seq.

With the aim of improving the state-of-the-art (SOTA) on a target task, a standard strategy in Natural Language Processing (NLP) research is to design a new model, or modify the existing SOTA, and then benchmark its performance on the target task. We argue in favor of enriching this chain of actions by a preliminary error-guided analysis: First, explore weaknesses by analyzing the hard cases where the existing model fails, and then target the improvement based on those. Interpretable evaluation has received little attention for structured prediction tasks. Therefore we propose the first in-depth analysis suite for Relation Classification (RC), and show its effectiveness through a case study. We propose a set of potentially influential attributes to focus on (e.g., entity distance, sentence length). Then, we bucket our datasets based on these attributes, and weight the importance of them through correlations. This allows us to identify highly challenging scenarios for the RC model. By exploiting the findings of our analysis, with a carefully targeted adjustment to our architecture, we effectively improve the performance over the baseline by >3 Micro-F1.

The labor market is changing rapidly, prompting increased interest in the automatic extraction of occupational skills from text. With the advent of English benchmark job description datasets, there is a need for systems that handle their diversity well. We tackle the complexity in occupational skill datasets tasks—combining and leveraging multiple datasets for skill extraction, to identify rarely observed skills within a dataset, and overcoming the scarcity of skills across datasets. In particular, we investigate the retrieval-augmentation of language models, employing an external datastore for retrieving similar skills in a dataset-unifying manner. Our proposed method, Nearest Neighbor Occupational Skill Extraction (NNOSE) effectively leverages multiple datasets by retrieving neighboring skills from other datasets in the datastore. This improves skill extraction without additional fine-tuning. Crucially, we observe a performance gain in predicting infrequent patterns, with substantial gains of up to 30% span-F1 in cross-dataset settings.

Recent years have brought significant advances to Natural Language Processing (NLP), which enabled fast progress in the field of computational job market analysis. Core tasks in this application domain are skill extraction and classification from job postings. Because of its quick growth and its interdisciplinary nature, there is no exhaustive assessment of this field. This survey aims to fill this gap by providing a comprehensive overview of deep learning methodologies, datasets, and terminologies specific to NLP-driven skill extraction. Our comprehensive cataloging of publicly available datasets addresses the lack of consolidated information on dataset creation and characteristics. Finally, the focus on terminology addresses the current lack of consistent definitions for important concepts, such as hard and soft skills, and terms relating to skill extraction and classification.

Large language models have emerged as a useful technology for job matching, for both candidates and employers. Job matching is often based on a particular geographic location, such as a city or region. However, LMs have known biases, commonly derived from their training data. In this work, we aim to quantify the metropolitan size bias encoded within large language models, evaluating zero-shot salary, employer presence, and commute duration predictions in 384 of the United States’ metropolitan regions. Across all benchmarks, we observe correlations between metropolitan population and the accuracy of predictions, with the smallest 10 metropolitan regions showing upwards of 300% worse benchmark performance than the largest 10.

Textual domain is a crucial property within the Natural Language Processing (NLP) community due to its effects on downstream model performance. The concept itself is, however, loosely defined and, in practice, refers to any non-typological property, such as genre, topic, medium or style of a document. We investigate the core notion of domains via human proficiency in identifying related intrinsic textual properties, specifically the concepts of genre (communicative purpose) and topic (subject matter). We publish our annotations in TGeGUM: A collection of 9.1k sentences from the GUM dataset (Zeldes, 2017) with single sentence and larger context (i.e., prose) annotations for one of 11 genres (source type), and its topic/subtopic as per the Dewey Decimal library classification system (Dewey, 1979), consisting of 10/100 hierarchical topics of increased granularity. Each instance is annotated by three annotators, for a total of 32.7k annotations, allowing us to examine the level of human disagreement and the relative difficulty of each annotation task. With a Fleiss’ kappa of at most 0.53 on the sentence level and 0.66 at the prose level, it is evident that despite the ubiquity of domains in NLP, there is little human consensus on how to define them. By training classifiers to perform the same task, we find that this uncertainty also extends to NLP models.

When creating a new dataset for evaluation, one of the first considerations is the size of the dataset. If our evaluation data is too small, we risk making unsupported claims based on the results on such data. If, on the other hand, the data is too large, we waste valuable annotation time and costs that could have been used to widen the scope of our evaluation (i.e. annotate for more domains/languages). Hence, we investigate the effect of the size and a variety of sampling strategies of evaluation data to optimize annotation efforts, using dependency parsing as a test case. We show that for in-language in-domain datasets, 5,000 tokens is enough to obtain a reliable ranking of different parsers; especially if the data is distant enough from the training split (otherwise, we recommend 10,000). In cross-domain setups, the same amounts are required, but in cross-lingual setups much less (2,000 tokens) is enough.

Current language models require a lot of training data to obtain high performance. For Relation Classification (RC), many datasets are domain-specific, so combining datasets to obtain better performance is non-trivial. We explore a multi-domain training setup for RC, and attempt to improve performance by encoding domain information. Our proposed models improve > 2 Macro-F1 against the baseline setup, and our analysis reveals that not all the labels benefit the same: The classes which occupy a similar space across domains (i.e., their interpretation is close across them, for example “physical”) benefit the least, while domain-dependent relations (e.g., “part-of”) improve the most when encoding domain information.

Digital assistants perform well in high-resource languages like English, where tasks like slot and intent detection (SID) are well-supported. Many recent SID datasets start including multiple language varieties. However, it is unclear how realistic these translated datasets are. Therefore, we extend one such dataset, namely xSID-0.4, to include two underrepresented languages: Bavarian, a German dialect, and Lithuanian, a Baltic language. Both language variants have limited speaker populations and are often not included in multilingual projects. In addition to translations we provide “natural” queries to digital assistants generated by native speakers. We further include utterances from another dataset for Bavarian to build the richest SID dataset available today for a low-resource dialect without standard orthography. We then set out to evaluate models trained on English in a zero-shot scenario on our target language variants. Our evaluation reveals that translated data can produce overly optimistic scores. However, the error patterns in translated and natural datasets are highly similar. Cross-dataset experiments demonstrate that data collection methods influence performance, with scores lower than those achieved with single-dataset translations. This work contributes to enhancing SID datasets for underrepresented languages, yielding NaLiBaSID, a new evaluation dataset for Bavarian and Lithuanian.

To improve the ability of language models to handle Natural Language Processing(NLP) tasks and intermediate step of pre-training has recently beenintroduced. In this setup, one takes a pre-trained language model, trains it ona (set of) NLP dataset(s), and then finetunes it for a target task. It isknown that the selection of relevant transfer tasks is important, but recentlysome work has shown substantial performance gains by doing intermediatetraining on a very large set of datasets. Most previous work uses generativelanguage models or only focuses on one or a couple of tasks and uses acarefully curated setup. We compare intermediate training with one or manytasks in a setup where the choice of datasets is more arbitrary; we use allSemEval 2023 text-based tasks. We reach performance improvements for most taskswhen using intermediate training. Gains are higher when doing intermediatetraining on single tasks than all tasks if the right transfer taskis identified. Dataset smoothing and heterogeneous batching did not lead torobust gains in our setup.

The increasing number of benchmarks for Natural Language Processing (NLP) tasks in the computational job market domain highlights the demand for methods that can handle job-related tasks such as skill extraction, skill classification, job title classification, and de-identification. While some approaches have been developed that are specific to the job market domain, there is a lack of generalized, multilingual models and benchmarks for these tasks. In this study, we introduce a language model called ESCOXLM-R, based on XLM-R-large, which uses domain-adaptive pre-training on the European Skills, Competences, Qualifications and Occupations (ESCO) taxonomy, covering 27 languages. The pre-training objectives for ESCOXLM-R include dynamic masked language modeling and a novel additional objective for inducing multilingual taxonomical ESCO relations. We comprehensively evaluate the performance of ESCOXLM-R on 6 sequence labeling and 3 classification tasks in 4 languages and find that it achieves state-of-the-art results on 6 out of 9 datasets. Our analysis reveals that ESCOXLM-R performs better on short spans and outperforms XLM-R-large on entity-level and surface-level span-F1, likely due to ESCO containing short skill and occupation titles, and encoding information on the entity-level.

Numerous studies found that the linguistic properties of a person’s native language affect the cognitive processing of other languages. However, only one study has shown that it was possible to identify the native language based on eye-tracking records of natural L2 reading using machine learning. A new corpus allows us to replicate these results on a more interrelated and larger set of native languages. Our results show that comparable classification performance is maintained despite using less data. However, analysis shows that the correlation between L2 eye movements and native language similarity may be more complex than the original study found.

Relation Extraction (RE) remains a challenging task, especially when considering realistic out-of-domain evaluations. One of the main reasons for this is the limited training size of current RE datasets: obtaining high-quality (manually annotated) data is extremely expensive and cannot realistically be repeated for each new domain. An intermediate training step on data from related tasks has shown to be beneficial across many NLP tasks. However, this setup still requires supplementary annotated data, which is often not available. In this paper, we investigate intermediate pre-training specifically for RE. We exploit the affinity between syntactic structure and semantic RE, and identify the syntactic relations which are closely related to RE by being on the shortest dependency path between two entities. We then take advantage of the high accuracy of current syntactic parsers in order to automatically obtain large amounts of low-cost pre-training data. By pre-training our RE model on the relevant syntactic relations, we are able to outperform the baseline in five out of six cross-domain setups, without any additional annotated data.

Representational spaces learned via language modeling are fundamental to Natural Language Processing (NLP), however there has been limited understanding regarding how and when during training various types of linguistic information emerge and interact. Leveraging a novel information theoretic probing suite, which enables direct comparisons of not just task performance, but their representational subspaces, we analyze nine tasks covering syntax, semantics and reasoning, across 2M pre-training steps and five seeds. We identify critical learning phases across tasks and time, during which subspaces emerge, share information, and later disentangle to specialize. Across these phases, syntactic knowledge is acquired rapidly after 0.5% of full training. Continued performance improvements primarily stem from the acquisition of open-domain knowledge, while semantics and reasoning tasks benefit from later boosts to long-range contextualization and higher specialization. Measuring cross-task similarity further reveals that linguistically related tasks share information throughout training, and do so more during the critical phase of learning than before or after. Our findings have implications for model interpretability, multi-task learning, and learning from limited data.

Language understanding is a multi-faceted cognitive capability, which the Natural Language Processing (NLP) community has striven to model computationally for decades. Traditionally, facets of linguistic intelligence have been compartmentalized into tasks with specialized model architectures and corresponding evaluation protocols. With the advent of large language models (LLMs) the community has witnessed a dramatic shift towards general purpose, task-agnostic approaches powered by generative models. As a consequence, the traditional compartmentalized notion of language tasks is breaking down, followed by an increasing challenge for evaluation and analysis. At the same time, LLMs are being deployed in more real-world scenarios, including previously unforeseen zero-shot setups, increasing the need for trustworthy and reliable systems. Therefore, we argue that it is time to rethink what constitutes tasks and model evaluation in NLP, and pursue a more holistic view on language, placing trustworthiness at the center. Towards this goal, we review existing compartmentalized approaches for understanding the origins of a model’s functional capacity, and provide recommendations for more multi-faceted evaluation protocols.

This report presents the results of the shared tasks organized as part of the VarDial Evaluation Campaign 2023. The campaign is part of the tenth workshop on Natural Language Processing (NLP) for Similar Languages, Varieties and Dialects (VarDial), co-located with EACL 2023. Three separate shared tasks were included this year: Slot and intent detection for low-resource language varieties (SID4LR), Discriminating Between Similar Languages – True Labels (DSL-TL), and Discriminating Between Similar Languages – Speech (DSL-S). All three tasks were organized for the first time this year.

Most research in Relation Extraction (RE) involves the English language, mainly due to the lack of multi-lingual resources. We propose Multi-CrossRE, the broadest multi-lingual dataset for RE, including 26 languages in addition to English, and covering six text domains. Multi-CrossRE is a machine translated version of CrossRE (Bassignana and Plank, 2022), with a sub-portion including more than 200 sentences in seven diverse languages checked by native speakers. We run a baseline model over the 26 new datasets and–as sanity check–over the 26 back-translations to English. Results on the back-translated data are consistent with the ones on the original English CrossRE, indicating high quality of the translation and the resulting dataset.

Language from social media remains challenging to process automatically, especially for non-English languages. In this work, we introduce the first NLP dataset for TikTok comments and the first Danish social media dataset with part-of-speech annotation. We further supply annotations for normalization, code-switching, and annotator uncertainty. As transferring models to such a highly specialized domain is non-trivial, we conduct an extensive study into which source data and modeling decisions most impact the performance. Surprisingly, transferring from in-domain data, even from a different language, outperforms in-language, out-of-domain training. These benefits nonetheless rely on the underlying language models having been at least partially pre-trained on data from the target language. Using our additional annotation layers, we further analyze how normalization, code-switching, and human uncertainty affect the tagging accuracy.

Cross-lingual transfer of parsing models has been shown to work well for several closely-related languages, but predicting the success in other cases remains hard. Our study is a comprehensive analysis of the impact of linguistic distance on the transfer of UD parsers. As an alternative to syntactic typological distances extracted from URIEL, we propose three text-based feature spaces and show that they can be more precise predictors, especially on a more local scale, when only shorter distances are taken into account. Our analyses also reveal that the good coverage in typological databases is not among the factors that explain good transfer.

Making an informed choice of pre-trained language model (LM) is critical for performance, yet environmentally costly, and as such widely underexplored. The field of Computer Vision has begun to tackle encoder ranking, with promising forays into Natural Language Processing, however they lack coverage of linguistic tasks such as structured prediction. We propose probing to rank LMs, specifically for parsing dependencies in a given language, by measuring the degree to which labeled trees are recoverable from an LM’s contextualized embeddings. Across 46 typologically and architecturally diverse LM-language pairs, our probing approach predicts the best LM choice 79% of the time using orders of magnitude less compute than training a full parser. Within this study, we identify and analyze one recently proposed decoupled LM—RemBERT—and find it strikingly contains less inherent dependency information, but often yields the best parser after full fine-tuning. Without this outlier our approach identifies the best LM in 89% of cases.

Probing has become an important tool for analyzing representations in Natural Language Processing (NLP). For graphical NLP tasks such as dependency parsing, linear probes are currently limited to extracting undirected or unlabeled parse trees which do not capture the full task. This work introduces DepProbe, a linear probe which can extract labeled and directed dependency parse trees from embeddings while using fewer parameters and compute than prior methods. Leveraging its full task coverage and lightweight parametrization, we investigate its predictive power for selecting the best transfer language for training a full biaffine attention parser. Across 13 languages, our proposed method identifies the best source treebank 94% of the time, outperforming competitive baselines and prior work. Finally, we analyze the informativeness of task-specific subspaces in contextual embeddings as well as which benefits a full parser’s non-linear parametrization provides.

Linguistic information is encoded at varying timescales (subwords, phrases, etc.) and communicative levels, such as syntax and semantics. Contextualized embeddings have analogously been found to capture these phenomena at distinctive layers and frequencies. Leveraging these findings, we develop a fully learnable frequency filter to identify spectral profiles for any given task. It enables vastly more granular analyses than prior handcrafted filters, and improves on efficiency. After demonstrating the informativeness of spectral probing over manual filters in a monolingual setting, we investigate its multilingual characteristics across seven diverse NLP tasks in six languages. Our analyses identify distinctive spectral profiles which quantify cross-task similarity in a linguistically intuitive manner, while remaining consistent across languages—highlighting their potential as robust, lightweight task descriptors.

The field of Deep Learning (DL) has undergone explosive growth during the last decade, with a substantial impact on Natural Language Processing (NLP) as well. Yet, compared to more established disciplines, a lack of common experimental standards remains an open challenge to the field at large. Starting from fundamental scientific principles, we distill ongoing discussions on experimental standards in NLP into a single, widely-applicable methodology. Following these best practices is crucial to strengthen experimental evidence, improve reproducibility and enable scientific progress. These standards are further collected in a public repository to help them transparently adapt to future needs.

As input representation for each sub-word, the original BERT architecture proposes the sum of the sub-word embedding, position embedding and a segment embedding. Sub-word and position embeddings are well-known and studied, and encode lexical information and word position, respectively. In contrast, segment embeddings are less known and have so far received no attention, despite being ubiquitous in large pre-trained language models. The key idea of segment embeddings is to encode to which of the two sentences (segments) a word belongs to — the intuition is to inform the model about the separation of sentences for the next sentence prediction pre-training task. However, little is known on whether the choice of segment impacts performance. In this work, we try to fill this gap and empirically study the impact of the segment embedding during inference time for a variety of pre-trained embeddings and target tasks. We hypothesize that for single-sentence prediction tasks performance is not affected — neither in mono- nor multilingual setups — while it matters when swapping segment IDs in paired-sentence tasks. To our surprise, this is not the case. Although for classification tasks and monolingual BERT models no large differences are observed, particularly word-level multilingual prediction tasks are heavily impacted. For low-resource syntactic tasks, we observe impacts of segment embedding and multilingual BERT choice. We find that the default setting for the most used multilingual BERT model underperforms heavily, and a simple swap of the segment embeddings yields an average improvement of 2.5 points absolute LAS score for dependency parsing over 9 different treebanks.

Automatically detecting the intent of an utterance is important for various downstream natural language processing tasks. This task is also called Dialogue Act Classification (DAC) and was primarily researched on spoken one-to-one conversations. The rise of social media has made this an interesting data source to explore within DAC, although it comes with some difficulties: non-standard form, variety of language types (across and within platforms), and quickly evolving norms. We therefore investigate the robustness of DAC on social media data in this paper. More concretely, we provide a benchmark that includes cross-domain data splits, as well as a variety of improvements on our transformer-based baseline. Our experiments show that lexical normalization is not beneficial in this setup, balancing the labels through resampling is beneficial in some cases, and incorporating context is crucial for this task and leads to the highest performance improvements 7 F1 percentage points in-domain and 20 cross-domain).

Previous work on multi-task learning in Natural Language Processing (NLP) oftenincorporated carefully selected tasks as well as carefully tuning ofarchitectures to share information across tasks. Recently, it has shown thatfor autoregressive language models, a multi-task second pre-training step on awide variety of NLP tasks leads to a set of parameters that more easily adaptfor other NLP tasks. In this paper, we examine whether a similar setup can beused in autoencoder language models using a restricted set of semanticallyoriented NLP tasks, namely all SemEval 2022 tasks that are annotated at theword, sentence or paragraph level. We first evaluate a multi-task model trainedon all SemEval 2022 tasks that contain annotation on the word, sentence orparagraph level (7 tasks, 11 sub-tasks), and then evaluate whetherre-finetuning the resulting model for each task specificially leads to furtherimprovements. Our results show that our mono-task baseline, our multi-taskmodel and our re-finetuned multi-task model each outperform the other modelsfor a subset of the tasks. Overall, huge gains can be observed by doingmulti-task learning: for three tasks we observe an error reduction of more than40%.

Various historical languages, which used to be lingua franca of science and arts, deserve the attention of current NLP research. In this work, we take the first data-driven steps towards this research line for Classical Arabic (CA) by addressing named entity recognition (NER) and topic modeling (TM) on the example of CA literature. We manually annotate the encyclopedic work of Tafsir Al-Tabari with span-based NEs, sentence-based topics, and span-based subtopics, thus creating the Tafsir Dataset with over 51,000 sentences, the first large-scale multi-task benchmark for CA. Next, we analyze our newly generated dataset, which we make open-source available, with current language models (lightweight BiLSTM, transformer-based MaChAmP) along a novel script compression method, thereby achieving state-of-the-art performance for our target task CA-NER. We also show that CA-TM from the perspective of historical topic models, which are central to Arabic studies, is very challenging. With this interdisciplinary work, we lay the foundations for future research on automatic analysis of CA literature.

Lexical normalization is the task of transforming an utterance into its standardized form. This task is beneficial for downstream analysis, as it provides a way to harmonize (often spontaneous) linguistic variation. Such variation is typical for social media on which information is shared in a multitude of ways, including diverse languages and code-switching. Since the seminal work of Han and Baldwin (2011) a decade ago, lexical normalization has attracted attention in English and multiple other languages. However, there exists a lack of a common benchmark for comparison of systems across languages with a homogeneous data and evaluation setup. The MultiLexNorm shared task sets out to fill this gap. We provide the largest publicly available multilingual lexical normalization benchmark including 13 language variants. We propose a homogenized evaluation setup with both intrinsic and extrinsic evaluation. As extrinsic evaluation, we use dependency parsing and part-of-speech tagging with adapted evaluation metrics (a-LAS, a-UAS, and a-POS) to account for alignment discrepancies. The shared task hosted at W-NUT 2021 attracted 9 participants and 18 submissions. The results show that neural normalization systems outperform the previous state-of-the-art system by a large margin. Downstream parsing and part-of-speech tagging performance is positively affected but to varying degrees, with improvements of up to 1.72 a-LAS, 0.85 a-UAS, and 1.54 a-POS for the winning system.

Social media is notoriously difficult to process for existing natural language processing tools, because of spelling errors, non-standard words, shortenings, non-standard capitalization and punctuation. One method to circumvent these issues is to normalize input data before processing. Most previous work has focused on only one language, which is mostly English. In this paper, we are the first to propose a model for cross-lingual normalization, with which we participate in the WNUT 2021 shared task. To this end, we use MoNoise as a starting point, and make a simple adaptation for cross-lingual application. Our proposed model outperforms the leave-as-is baseline provided by the organizers which copies the input. Furthermore, we explore a completely different model which converts the task to a sequence labeling task. Performance of this second system is low, as it does not take capitalization into account in our implementation.

The lack of publicly available evaluation data for low-resource languages limits progress in Spoken Language Understanding (SLU). As key tasks like intent classification and slot filling require abundant training data, it is desirable to reuse existing data in high-resource languages to develop models for low-resource scenarios. We introduce xSID, a new benchmark for cross-lingual (x) Slot and Intent Detection in 13 languages from 6 language families, including a very low-resource dialect. To tackle the challenge, we propose a joint learning approach, with English SLU training data and non-English auxiliary tasks from raw text, syntax and translation for transfer. We study two setups which differ by type and language coverage of the pre-trained embeddings. Our results show that jointly learning the main tasks with masked language modeling is effective for slots, while machine translation transfer works best for intent classification.

Because of globalization, it is becoming more and more common to use multiple languages in a single utterance, also called code-switching. This results in special linguistic structures and, therefore, poses many challenges for Natural Language Processing. Existing models for language identification in code-switched data are all supervised, requiring annotated training data which is only available for a limited number of language pairs. In this paper, we explore semi-supervised approaches, that exploit out-of-domain mono-lingual training data. We experiment with word uni-grams, word n-grams, character n-grams, Viterbi Decoding, Latent Dirichlet Allocation, Support Vector Machine and Logistic Regression. The Viterbi model was the best semi-supervised model, scoring a weighted F1 score of 92.23%, whereas a fully supervised state-of-the-art BERT-based model scored 98.43%.

Lexical normalization, the translation of non-canonical data to standard language, has shown to improve the performance of many natural language processing tasks on social media. Yet, using multiple languages in one utterance, also called code-switching (CS), is frequently overlooked by these normalization systems, despite its common use in social media. In this paper, we propose three normalization models specifically designed to handle code-switched data which we evaluate for two language pairs: Indonesian-English and Turkish-German. For the latter, we introduce novel normalization layers and their corresponding language ID and POS tags for the dataset, and evaluate the downstream effect of normalization on POS tagging. Results show that our CS-tailored normalization models significantly outperform monolingual ones, and lead to 5.4% relative performance increase for POS tagging as compared to unnormalized input.

Transfer learning, particularly approaches that combine multi-task learning with pre-trained contextualized embeddings and fine-tuning, have advanced the field of Natural Language Processing tremendously in recent years. In this paper we present MaChAmp, a toolkit for easy fine-tuning of contextualized embeddings in multi-task settings. The benefits of MaChAmp are its flexible configuration options, and the support of a variety of natural language processing tasks in a uniform toolkit, from text classification and sequence labeling to dependency parsing, masked language modeling, and text generation.

While high performance have been obtained for high-resource languages, performance on low-resource languages lags behind. In this paper we focus on the parsing of the low-resource language Frisian. We use a sample of code-switched, spontaneously spoken data, which proves to be a challenging setup. We propose to train a parser specifically tailored towards the target domain, by selecting instances from multiple treebanks. Specifically, we use Latent Dirichlet Allocation (LDA), with word and character N-grams. We use a deep biaffine parser initialized with mBERT. The best single source treebank (nl_alpino) resulted in an LAS of 54.7 whereas our data selection outperformed the single best transfer treebank and led to 55.6 LAS on the test data. Additional experiments consisted of removing diacritics from our Frisian data, creating more similar training data by cropping sentences and running our best model using XLM-R. These experiments did not lead to a better performance.

Recent complementary strands of research have shown that leveraging information on the data source through encoding their properties into embeddings can lead to performance increase when training a single model on heterogeneous data sources. However, it remains unclear in which situations these dataset embeddings are most effective, because they are used in a large variety of settings, languages and tasks. Furthermore, it is usually assumed that gold information on the data source is available, and that the test data is from a distribution seen during training. In this work, we compare the effect of dataset embeddings in mono-lingual settings, multi-lingual settings, and with predicted data source label in a zero-shot setting. We evaluate on three morphosyntactic tasks: morphological tagging, lemmatization, and dependency parsing, and use 104 datasets, 66 languages, and two different dataset grouping strategies. Performance increases are highest when the datasets are of the same language, and we know from which distribution the test-instance is drawn. In contrast, for setups where the data is from an unseen distribution, performance increase vanishes.

Standard train-dev-test splits used to benchmark multiple models against each other are ubiquitously used in Natural Language Processing (NLP). In this setup, the train data is used for training the model, the development set for evaluating different versions of the proposed model(s) during development, and the test set to confirm the answers to the main research question(s). However, the introduction of neural networks in NLP has led to a different use of these standard splits; the development set is now often used for model selection during the training procedure. Because of this, comparing multiple versions of the same model during development leads to overestimation on the development data. As an effect, people have started to compare an increasing amount of models on the test data, leading to faster overfitting and “expiration” of our test sets. We propose to use a tune-set when developing neural network methods, which can be used for model picking so that comparing the different versions of a new model can safely be done on the development data.

Recent work has shown that monolingual masked language models learn to represent data-driven notions of language variation which can be used for domain-targeted training data selection. Dataset genre labels are already frequently available, yet remain largely unexplored in cross-lingual setups. We harness this genre metadata as a weak supervision signal for targeted data selection in zero-shot dependency parsing. Specifically, we project treebank-level genre information to the finer-grained sentence level, with the goal to amplify information implicitly stored in unsupervised contextualized representations. We demonstrate that genre is recoverable from multilingual contextual embeddings and that it provides an effective signal for training data selection in cross-lingual, zero-shot scenarios. For 12 low-resource language treebanks, six of which are test-only, our genre-specific methods significantly outperform competitive baselines as well as recent embedding-based methods for data selection. Moreover, genre-based data selection provides new state-of-the-art results for three of these target languages.

This paper introduces DAN+, a new multi-domain corpus and annotation guidelines for Dan-ish nested named entities (NEs) and lexical normalization to support research on cross-lingualcross-domain learning for a less-resourced language. We empirically assess three strategies tomodel the two-layer Named Entity Recognition (NER) task. We compare transfer capabilitiesfrom German versus in-language annotation from scratch. We examine language-specific versusmultilingual BERT, and study the effect of lexical normalization on NER. Our results show that 1) the most robust strategy is multi-task learning which is rivaled by multi-label decoding, 2) BERT-based NER models are sensitive to domain shifts, and 3) in-language BERT and lexicalnormalization are the most beneficial on the least canonical data. Our results also show that anout-of-domain setup remains challenging, while performance on news plateaus quickly. Thishighlights the importance of cross-domain evaluation of cross-lingual transfer.

With the COVID-19 pandemic raging world-wide since the beginning of the 2020 decade, the need for monitoring systems to track relevant information on social media is vitally important. This paper describes our submission to the WNUT-2020 Task 2: Identification of informative COVID-19 English Tweets. We investigate the effectiveness for a variety of classification models, and found that domain-specific pre-trained BERT models lead to the best performance. On top of this, we attempt a variety of ensembling strategies, but these attempts did not lead to further improvements. Our final best model, the standalone CT-BERT model, proved to be highly competitive, leading to a shared first place in the shared task. Our results emphasize the importance of domain and task-related pre-training.

Lexical normalization is the task of translating non-standard social media data to a standard form. Previous work has shown that this is beneficial for many downstream tasks in multiple languages. However, for Italian, there is no benchmark available for lexical normalization, despite the presence of many benchmarks for other tasks involving social media data. In this paper, we discuss the creation of a lexical normalization dataset for Italian. After two rounds of annotation, a Cohen’s kappa score of 78.64 is obtained. During this process, we also analyze the inter-annotator agreement for this task, which is only rarely done on datasets for lexical normalization,and when it is reported, the analysis usually remains shallow. Furthermore, we utilize this dataset to train a lexical normalization model and show that it can be used to improve dependency parsing of social media data. All annotated data and the code to reproduce the results are available at: http://bitbucket.org/robvanderg/normit.

Social media is a valuable data resource for various natural language processing (NLP) tasks. However, standard NLP tools were often designed with standard texts in mind, and their performance decreases heavily when applied to social media data. One solution to this problem is to adapt the input text to a more standard form, a task also referred to as normalization. Automatic approaches to normalization have shown that they can be used to improve performance on a variety of NLP tasks. However, all of these systems are supervised, thereby being heavily dependent on the availability of training data for the correct language and domain. In this work, we attempt to overcome this dependence by automatically generating training data for lexical normalization. Starting with raw tweets, we attempt two directions, to insert non-standardness (noise) and to automatically normalize in an unsupervised setting. Our best results are achieved by automatically inserting noise. We evaluate our approaches by using an existing lexical normalization system; our best scores are achieved by custom error generation system, which makes use of some manually created datasets. With this system, we score 94.29 accuracy on the test data, compared to 95.22 when it is trained on human-annotated data. Our best system which does not depend on any type of annotation is based on word embeddings and scores 92.04 accuracy. Finally, we perform an experiment in which we asked humans to predict whether a sentence was written by a human or generated by our best model. This experiment showed that in most cases it is hard for a human to detect automatically generated sentences.

Analogies such as man is to king as woman is to X are often used to illustrate the amazing power of word embeddings. Concurrently, they have also been used to expose how strongly human biases are encoded in vector spaces trained on natural language, with examples like man is to computer programmer as woman is to homemaker. Recent work has shown that analogies are in fact not an accurate diagnostic for bias, but this does not mean that they are not used anymore, or that their legacy is fading. Instead of focusing on the intrinsic problems of the analogy task as a bias detection tool, we discuss a series of issues involving implementation as well as subjective choices that might have yielded a distorted picture of bias in word embeddings. We stand by the truth that human biases are present in word embeddings, and, of course, the need to address them. But analogies are not an accurate tool to do so, and the way they have been most often used has exacerbated some possibly non-existing biases and perhaps hidden others. Because they are still widely popular, and some of them have become classics within and outside the NLP community, we deem it important to provide a series of clarifications that should put well-known, and potentially new analogies, into the right perspective.

We introduce Biomedical Event Extraction as Sequence Labeling (BeeSL), a joint end-to-end neural information extraction model. BeeSL recasts the task as sequence labeling, taking advantage of a multi-label aware encoding strategy and jointly modeling the intermediate tasks via multi-task learning. BeeSL is fast, accurate, end-to-end, and unlike current methods does not require any external knowledge base or preprocessing tools. BeeSL outperforms the current best system (Li et al., 2019) on the Genia 2011 benchmark by 1.57% absolute F1 score reaching 60.22% F1, establishing a new state of the art for the task. Importantly, we also provide first results on biomedical event extraction without gold entity information. Empirical results show that BeeSL’s speed and accuracy makes it a viable approach for large-scale real-world scenarios.

In this paper, we present our approach to detection of hate speech against women and immigrants in tweets for our participation in the SemEval-2019 Task 5. We trained an SVM and an RF classifier using character bi- and trigram features and a BiLSTM pre-initialized with external word embeddings. We combined the predictions of the SVM, RF and BiLSTM in two different ensemble models. The first was a majority vote of the binary values, and the second used the average of the confidence scores. For development, we got the highest accuracy (75%) by the final ensemble model with majority voting. For testing, all models scored substantially lower and the scores between the classifiers varied more. We believe that these large differences between the higher accuracies in the development phase and the lower accuracies we obtained in the testing phase have partly to do with differences between the training, development and testing data.

In this paper, we introduce and demonstrate the online demo as well as the command line interface of a lexical normalization system (MoNoise) for a variety of languages. We further improve this model by using features from the original word for every normalization candidate. For comparison with future work, we propose the bundling of seven datasets in six languages to form a new benchmark, together with a novel evaluation metric which is particularly suitable for cross-dataset comparisons. MoNoise reaches a new state-of-art performance for six out of seven of these datasets. Furthermore, we allow the user to tune the ‘aggressiveness’ of the normalization, and show how the model can be made more efficient with only a small loss in performance. The online demo can be found on: http://www.robvandergoot.com/monoise and the corresponding code on: https://bitbucket.org/robvanderg/monoise/

Existing natural language processing systems have often been designed with standard texts in mind. However, when these tools are used on the substantially different texts from social media, their performance drops dramatically. One solution is to translate social media data to standard language before processing, this is also called normalization. It is well-known that this improves performance for many natural language processing tasks on social media data. However, little is known about which types of normalization replacements have the most effect. Furthermore, it is unknown what the weaknesses of existing lexical normalization systems are in an extrinsic setting. In this paper, we analyze the effect of manual as well as automatic lexical normalization for dependency parsing. After our analysis, we conclude that for most categories, automatic normalization scores close to manually annotated normalization and that small annotation differences are important to take into consideration when exploiting normalization in a pipeline setup.

This paper describes our submission to SIGMORPHON 2019 Task 2: Morphological analysis and lemmatization in context. Our model is a multi-task sequence to sequence neural network, which jointly learns morphological tagging and lemmatization. On the encoding side, we exploit character-level as well as contextual information. We introduce a multi-attention decoder to selectively focus on different parts of character and word sequences. To further improve the model, we train on multiple datasets simultaneously and use external embeddings for initialization. Our final model reaches an average morphological tagging F1 score of 94.54 and a lemma accuracy of 93.91 on the test data, ranking respectively 3rd and 6th out of 13 teams in the SIGMORPHON 2019 shared task.

Gender prediction has typically focused on lexical and social network features, yielding good performance, but making systems highly language-, topic-, and platform dependent. Cross-lingual embeddings circumvent some of these limitations, but capture gender-specific style less. We propose an alternative: bleaching text, i.e., transforming lexical strings into more abstract features. This study provides evidence that such features allow for better transfer across languages. Moreover, we present a first study on the ability of humans to perform cross-lingual gender prediction. We find that human predictive power proves similar to that of our bleached models, and both perform better than lexical models.

Recently introduced neural network parsers allow for new approaches to circumvent data sparsity issues by modeling character level information and by exploiting raw data in a semi-supervised setting. Data sparsity is especially prevailing when transferring to non-standard domains. In this setting, lexical normalization has often been used in the past to circumvent data sparsity. In this paper, we investigate whether these new neural approaches provide similar functionality as lexical normalization, or whether they are complementary. We provide experimental results which show that a separate normalization component improves performance of a neural network parser even if it has access to character level information as well as external word embeddings. Further improvements are obtained by a straightforward but novel approach in which the top-N best candidates provided by the normalization component are available to the parser.

Does normalization help Part-of-Speech (POS) tagging accuracy on noisy, non-canonical data? To the best of our knowledge, little is known on the actual impact of normalization in a real-world scenario, where gold error detection is not available. We investigate the effect of automatic normalization on POS tagging of tweets. We also compare normalization to strategies that leverage large amounts of unlabeled data kept in its raw form. Our results show that normalization helps, but does not add consistently beyond just word embedding layer initialization. The latter approach yields a tagging model that is competitive with a Twitter state-of-the-art tagger.

This work explores different approaches of using normalization for parser adaptation. Traditionally, normalization is used as separate pre-processing step. We show that integrating the normalization model into the parsing algorithm is more beneficial. This way, multiple normalization candidates can be leveraged, which improves parsing performance on social media. We test this hypothesis by modifying the Berkeley parser; out-of-the-box it achieves an F1 score of 66.52. Our integrated approach reaches a significant improvement with an F1 score of 67.36, while using the best normalization sequence results in an F1 score of only 66.94.

We introduce the Denoised Web Treebank: a treebank including a normalization layer and a corresponding evaluation metric for dependency parsing of noisy text, such as Tweets. This benchmark enables the evaluation of parser robustness as well as text normalization methods, including normalization as machine translation and unsupervised lexical normalization, directly on syntactic trees. Experiments show that text normalization together with a combination of domain-specific and generic part-of-speech taggers can lead to a significant improvement in parsing accuracy on this test set.