Gregor Geigle, Florian Schneider, Carolin Holtermann, Chris Biemann, Radu Timofte, Anne Lauscher, Goran Glavaš. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Irina Nikishina, Saba Anwar, Nikolay Dolgov, Maria Manina, Daria Ignatenko, Artem Shelmanov, Chris Biemann. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Detoxifying offensive language while preserving the speaker's original intent is a challenging yet critical goal for improving the quality of online interactions.Although large language models (LLMs) show promise in rewriting toxic content, they often default to overly polite rewrites, distorting the emotional tone and communicative intent.This problem is especially acute in Chinese, where toxicity often arises implicitly through emojis, homophones, or discourse context.We present TOXIREWRITECN, the first Chinese detoxification dataset explicitly designed to preserve sentiment polarity.The dataset comprises 1,556 carefully annotated triplets, each containing a toxic sentence, a sentiment-aligned non-toxic rewrite, and labeled toxic spans.It covers five real-world scenarios: standard expressions, emoji-induced and homophonic toxicity, as well as single-turn and multi-turn dialogues.We evaluate 17 LLMs, including commercial and open-source models with variant architectures, across four dimensions: detoxification accuracy, fluency, content preservation, and sentiment polarity.Results show that while commercial and MoE models perform best overall, all models struggle to balance safety with emotional fidelity in more subtle or context-heavy settings such as emoji, homophone, and dialogue-based inputs.We release TOXIREWRITECN to support future research on controllable, sentiment-aware detoxification for Chinese.Caution: This paper contains examples of violent or offensive language that may be disturbing to some readers.

In this system demonstration paper, we describe the Whiteboards extension for an existing web-based platform for digital qualitative discourse analysis. Whiteboards comprise interactive graph-based interfaces to organize and manipulate objects, which can be qualitative research data, such as documents, images, etc., and analyses of these research data, such as annotations, tags, and code structures. The proposed extension offers a customizable view of the material and a wide range of actions that enable new ways of interacting and working with such resources. We show that the visualizations facilitate various use cases of qualitative data analysis, including reflection of the research process through sampling maps, creation of actor networks, and refining code taxonomies.

A wide body of research is concerned with the semantics of narratives, both in terms of understanding narratives and generating fictional narratives and stories. We provide a dataset of summaries to be used as a proxy for entire stories or for the analysis of the summaries themselves. Our dataset consists of a total of 96,831 individual summaries across 29,505 stories. We intend for the dataset to be used for training and evaluation of embedding representations for stories, specifically the stories’ narratives. The summary data is harvested from five different language versions of Wikipedia. Our dataset comes with rich metadata, which we extract from Wikidata, enabling a wide range of applications that operate on story summaries in conjunction with metadata. To set baseline results, we run retrieval experiments on the dataset, exploring the capability of similarity models in retrieving summaries of the same story. For this retrieval, a crucial element is to not place too much emphasis on the named entities, as this can enable retrieval of other summaries for the same work without taking the narrative into account.

Extracting who says what to whom is a crucial part in analyzing human communication in today’s abundance of data such as online news articles. Yet, the lack of annotated data for this task in German news articles severely limits the quality and usability of possible systems. To remedy this, we present a new, freely available, creative-commons-licensed dataset for quotation attribution in German news articles based on WIKINEWS. The dataset provides curated, high-quality annotations across 1000 documents (250,000 tokens) in a fine-grained annotation schema enabling various downstream uses for the dataset. The annotations not only specify who said what but also how, in which context, to whom and define the type of quotation. We specify our annotation schema, describe the creation of the dataset and provide a quantitative analysis. Further, we describe suitable evaluation metrics, apply two existing systems for quotation attribution, discuss their results to evaluate the utility of our dataset and outline use cases of our dataset in downstream tasks.

With the emergence of numerous Large Language Models (LLM), the usage of such models in various Natural Language Processing (NLP) applications is increasing extensively. Counterspeech generation is one such key task where efforts are made to develop generative models by fine-tuning LLMs with hatespeech - counterspeech pairs, but none of these attempts explores the intrinsic properties of large language models in zero-shot settings. In this work, we present a comprehensive analysis of the performances of four LLMs namely GPT-2, DialoGPT, ChatGPT and FlanT5 in zero-shot settings for counterspeech generation, which is the first of its kind. For GPT-2 and DialoGPT, we further investigate the deviation in performance with respect to the sizes (small, medium, large) of the models. On the other hand, we propose three different prompting strategies for generating different types of counterspeech and analyse the impact of such strategies on the performance of the models. Our analysis shows that there is an improvement in generation quality for two datasets (17%), however the toxicity increase (25%) with increase in model size. Considering type of model, GPT-2 and FlanT5 models are significantly better in terms of counterspeech quality but also have high toxicity as compared to DialoGPT. ChatGPT are much better at generating counter speech than other models across all metrics. In terms of prompting, we find that our proposed strategies help in improving counter speech generation across all the models.

Comparative Question Answering (CompQA) is a Natural Language Processing task that combines Question Answering and Argument Mining approaches to answer subjective comparative questions in an efficient argumentative manner. In this paper, we present an end-to-end (full pipeline) system for answering comparative questions called CAM 2.0 as well as a public leaderboard called CompUGE that unifies the existing datasets under a single easy-to-use evaluation suite. As compared to previous web-form-based CompQA systems, it features question identification, object and aspect labeling, stance classification, and summarization using up-to-date models. We also select the most time- and memory-effective pipeline by comparing separately fine-tuned Transformer Encoder models which show state-of-the-art performance on the subtasks with Generative LLMs in few-shot and LoRA setups. We also conduct a user study for a whole-system evaluation.

We present a novel approach to modeling fictional narratives.The proposed model creates embeddings that represent a story such that similar narratives, that is, reformulations of the same story, will result in similar embeddings.We showcase the prowess of our narrativefocused embeddings on various datasets, exhibiting state-of-the-art performance on multiple retrieval tasks.The embeddings also show promising results on a narrative understanding task.Additionally, we perform an annotationbased evaluation to validate that our introduced computational notion of narrative similarity aligns with human perception.The approach can help to explore vast datasets of stories, with potential applications in recommender systems and in the computational analysis of literature.

This paper paves the way for interpretable and configurable semantic similarity search, by training state-of-the-art models for identifying textual similarity guided by a set of aspects or dimensions. The similarity models are analyzed as to which interpretable dimensions of similarity they place the most emphasis on. We conceptually introduce configurable similarity search for finding documents similar in specific aspects but dissimilar in others. To evaluate the interpretability of these dimensions, we experiment with downstream retrieval tasks using weighted combinations of these dimensions. Configurable similarity search is an invaluable tool for exploring datasets and will certainly be helpful in many applied natural language processing research applications.

In this work, we present a self-supervised dual constraint contrastive method for efficiently fine-tuning the vision-language pre-trained (VLP) models that have achieved great success on various cross-modal tasks, since full fine-tune these pre-trained models is computationally expensive and tend to result in catastrophic forgetting restricted by the size and quality of labeled datasets. Our approach freezes the pre-trained VLP models as the fundamental, generalized, and transferable multimodal representation and incorporates lightweight parameters to learn domain and task-specific features without labeled data. We demonstrated that our self-supervised dual contrastive model performs better than previous fine-tuning methods on MS COCO and Flickr 30K datasets on the cross-modal retrieval task, with an even more pronounced improvement in zero-shot performance. Furthermore, experiments on the MOTIF dataset prove that our self-supervised approach remains effective when trained on a small, out-of-domain dataset without overfitting. As a plug-and-play method, our proposed method is agnostic to the underlying models and can be easily integrated with different VLP models, allowing for the potential incorporation of future advancements in VLP models.

Comparative Question Answering (cQA) is the task of providing concrete and accurate responses to queries such as: “Is Lyft cheaper than a regular taxi?” or “What makes a mortgage different from a regular loan?”. In this paper, we propose two new open-domain real-world datasets for identifying and labeling comparative questions. While the first dataset contains instances of English questions labeled as comparative vs. non-comparative, the second dataset provides additional labels including the objects and the aspects of comparison. We conduct several experiments that evaluate the soundness of our datasets. The evaluation of our datasets using various classifiers show promising results that reach close-to-human results on a binary classification task with a neural model using ALBERT embeddings. When approaching the unsupervised sequence labeling task, some headroom remains.

We present a dataset containing source code solutions to algorithmic programming exercises solved by hundreds of Bachelor-level students at the University of Hamburg. These solutions were collected during the winter semesters 2019/2020, 2020/2021 and 2021/2022. The dataset contains a set of solutions to a total of 21 tasks written in Java as well as Python and a total of over 1500 individual solutions. All solutions were submitted through Moodle and the Coderunner plugin and passed a number of test cases (including randomized tests), such that they can be considered as working correctly. All students whose solutions are included in the dataset gave their consent into publishing their solutions. The solutions are pseudonymized with a random solution ID. Included in this paper is a short analysis of the dataset containing statistical data and highlighting a few anomalies (e.g. the number of solutions per task decreases for the last few tasks due to grading rules). We plan to extend the dataset with tasks and solutions from upcoming courses.

MOTIF (MultimOdal ConTextualized Images For Language Learners) is a multimodal dataset that consists of 1125 comprehension texts retrieved from Wikipedia Simple Corpus. Allowing multimodal processing or enriching the context with multimodal information has proven imperative for many learning tasks, specifically for second language (L2) learning. In this respect, several traditional NLP approaches can assist L2 readers in text comprehension processes, such as simplifying text or giving dictionary descriptions for complex words. As nicely stated in the well-known proverb, sometimes “a picture is worth a thousand words” and an image can successfully complement the verbal message by enriching the representation, like in Pictionary books. This multimodal support can also assist on-the-fly text reading experience by providing a multimodal tool that chooses and displays the most relevant images for the difficult words, given the text context. This study mainly focuses on one of the key components to achieving this goal; collecting a multimodal dataset enriched with complex word annotation and validated image match.

In this work, we focus on the task of generating SPARQL queries from natural language questions, which can then be executed on Knowledge Graphs (KGs). We assume that gold entity and relations have been provided, and the remaining task is to arrange them in the right order along with SPARQL vocabulary, and input tokens to produce the correct SPARQL query. Pre-trained Language Models (PLMs) have not been explored in depth on this task so far, so we experiment with BART, T5 and PGNs (Pointer Generator Networks) with BERT embeddings, looking for new baselines in the PLM era for this task, on DBpedia and Wikidata KGs. We show that T5 requires special input tokenisation, but produces state of the art performance on LC-QuAD 1.0 and LC-QuAD 2.0 datasets, and outperforms task-specific models from previous works. Moreover, the methods enable semantic parsing for questions where a part of the input needs to be copied to the output query, thus enabling a new paradigm in KG semantic parsing.

In this work, we present the Golden Retriever, a system leveraging state-of-the-art visio-linguistic models (VLMs) for real-time text-image retrieval. The unique feature of our system is that it can focus on words contained in the textual query, i.e., locate and high-light them within retrieved images. An efficient two-stage process implements real-time capability and the ability to focus. Therefore, we first drastically reduce the number of images processed by a VLM. Then, in the second stage, we rank the images and highlight the focussed word using the outputs of a VLM. Further, we introduce a new and efficient algorithm based on the idea of TF-IDF to retrieve images for short textual queries. One of multiple use cases where we employ the Golden Retriever is a language learner scenario, where visual cues for "difficult" words within sentences are provided to improve a user's reading comprehension. However, since the backend is completely decoupled from the frontend, the system can be integrated into any other application where images must be retrieved fast. We demonstrate the Golden Retriever with screenshots of a minimalistic user interface.

Numerous attempts for hypernymy relation (e.g., dog “is-a” animal) detection have been made for resourceful languages like English, whereas efforts made for low-resource languages are scarce primarily due to lack of gold-standard datasets and suitable distributional models. Therefore, we introduce four gold-standard datasets for hypernymy detection for each of the two languages, namely, Hindi and Bengali, and two gold-standard datasets for Amharic. Another major contribution of this work is to prepare distributional thesaurus (DT) embeddings for all three languages using three different network embedding methods (DeepWalk, role2vec, and M-NMF) for the first time on these languages and to show their utility for hypernymy detection. Posing this problem as a binary classification task, we experiment with supervised classifiers like Support Vector Machine, Random Forest, and so on, and we show that these classifiers fed with DT embeddings can obtain promising results while evaluated against proposed gold-standard datasets, specifically in an experimental setup that counteracts lexical memorization. We further incorporate DT embeddings and pre-trained fastText embeddings together using two different hybrid approaches, both of which produce an excellent performance. Additionally, we validate our methodology on gold-standard English datasets as well, where we reach a comparable performance to state-of-the-art models for hypernymy detection.

Hate speech is a challenging issue plaguing the online social media. While better models for hate speech detection are continuously being developed, there is little research on the bias and interpretability aspects of hate speech. In this paper, we introduce HateXplain, the first benchmark hate speech dataset covering multiple aspects of the issue. Each post in our dataset is annotated from three different perspectives: the basic, commonly used 3-class classification (i.e., hate, offensive or normal), the target community (i.e., the community that has been the victim of hate speech/offensive speech in the post), and the rationales, i.e., the portions of the post on which their labelling decision (as hate, offensive or normal) is based. We utilize existing state-of-the-art models and observe that even models that perform very well in classification do not score high on explainability metrics like model plausibility and faithfulness. We also observe that models, which utilize the human rationales for training, perform better in reducing unintended bias towards target communities. We have made our code and dataset public for other researchers.

Reference resolution is an important problem that has enormous practical implications in daily life, for example in recovering the intended meaning in communication when the environment is noisy (acoustic noise in the spoken channel, or clutter / occlusion in the visual world). Recent literature indicates that cross-modal processing of all the contributive modalities improves the reference resolution in such settings. In this paper, we investigate the contribution of the eye-tracking methodology, a substantial but underrepresented component of face-to-face communication in NLP systems, to recover the meaning in noisy settings. We integrate gaze features into state-of-the-art language models and test the model on data where parts of the sentences are masked, mimicking noise in the acoustic channel. The results indicate that eye movements can compensate for the missing information in the situation and support communication when language and visual modality fail.

Sian Gooding, Ekaterina Kochmar, Seid Muhie Yimam, Chris Biemann. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Multi-task learning (MTL) and transfer learning (TL) are techniques to overcome the issue of data scarcity when training state-of-theart neural networks.However, finding beneficial auxiliary datasets for MTL or TL is a time-and resource-consuming trial-and-error approach.We propose new methods to automatically assess the similarity of sequence tagging datasets to identify beneficial auxiliary data for MTL or TL setups.Our methods can compute the similarity between any two sequence tagging datasets, i.e. they do not need to be annotated with the same tagset or multiple labels in parallel.Additionally, our methods take tokens and their labels into account, which is more robust than only using either of them as an information source, as conducted in prior work.We empirically show that our similarity measures correlate with the change in test score of neural networks that use the auxiliary dataset for MTL to increase the main task performance.We provide an efficient, opensource implementation. 1

This paper presents the study of sentiment analysis for Amharic social media texts.As the number of social media users is ever-increasing, social media platforms would like to understand the latent meaning and sentiments of a text to enhance decision-making procedures.However, lowresource languages such as Amharic have received less attention due to several reasons such as lack of well-annotated datasets, unavailability of computing resources, and fewer or no expert researchers in the area.This research addresses three main research questions.We first explore the suitability of existing tools for the sentiment analysis task.Annotation tools are scarce to support large-scale annotation tasks in Amharic.Also, the existing crowdsourcing platforms do not support Amharic text annotation.Hence, we build a social-network-friendly annotation tool called 'ASAB' using the Telegram bot.We collect 9.4k tweets, where each tweet is annotated by three Telegram users.Moreover, we explore the suitability of machine learning approaches for Amharic sentiment analysis.The FLAIR deep learning text classifier, based on network embeddings that are computed from a distributional thesaurus, outperforms other supervised classifiers.We further investigate the challenges in building a sentiment analysis system for Amharic and we found that the widespread usage of sarcasm and figurative speech are the main issues in dealing with the problem.To advance the sentiment analysis research in Amharic and other related lowresource languages, we release the dataset, the annotation tool, source code, and models publicly under a permissive.

The Sparsespeech model is an unsupervised acoustic model that can generate discrete pseudo-labels for untranscribed speech. We extend the Sparsespeech model to allow for sampling over a random discrete variable, yielding pseudo-posteriorgrams. The degree of sparsity in this posteriorgram can be fully controlled after the model has been trained. We use the Gumbel-Softmax trick to approximately sample from a discrete distribution in the neural network and this allows us to train the network efficiently with standard backpropagation. The new and improved model is trained and evaluated on the Libri-Light corpus, a benchmark for ASR with limited or no supervision. The model is trained on 600h and 6000h of English read speech. We evaluate the improved model using the ABX error measure and a semi-supervised setting with 10h of transcribed speech. We observe a relative improvement of up to 31.4% on ABX error rates across speakers on the test set with the improved Sparsespeech model on 600h of speech data and further improvements when we scale the model to 6000h.

Disambiguation of word senses in context is easy for humans, but is a major challenge for automatic approaches. Sophisticated supervised and knowledge-based models were developed to solve this task. However, (i) the inherent Zipfian distribution of supervised training instances for a given word and/or (ii) the quality of linguistic knowledge representations motivate the development of completely unsupervised and knowledge-free approaches to word sense disambiguation (WSD). They are particularly useful for under-resourced languages which do not have any resources for building either supervised and/or knowledge-based models. In this paper, we present a method that takes as input a standard pre-trained word embedding model and induces a fully-fledged word sense inventory, which can be used for disambiguation in context. We use this method to induce a collection of sense inventories for 158 languages on the basis of the original pre-trained fastText word embeddings by Grave et al., (2018), enabling WSD in these languages. Models and system are available online.

We present the first approach to automatically building resources for academic writing. The aim is to build a writing aid system that automatically edits a text so that it better adheres to the academic style of writing. On top of existing academic resources, such as the Corpus of Contemporary American English (COCA) academic Word List, the New Academic Word List, and the Academic Collocation List, we also explore how to dynamically build such resources that would be used to automatically identify informal or non-academic words or phrases. The resources are compiled using different generic approaches that can be extended for different domains and languages. We describe the evaluation of resources with a system implementation. The system consists of an informal word identification (IWI), academic candidate paraphrase generation, and paraphrase ranking components. To generate candidates and rank them in context, we have used the PPDB and WordNet paraphrase resources. We use the Concepts in Context (CoInCO) “All-Words” lexical substitution dataset both for the informal word identification and paraphrase generation experiments. Our informal word identification component achieves an F-1 score of 82%, significantly outperforming a stratified classifier baseline. The main contribution of this work is a domain-independent methodology to build targeted resources for writing aids.

\beginabstract We analyze comparative questions, i.e., questions asking to compare different items, that were submitted to Yandex in 2012. Responses to such questions might be quite different from the simple "ten blue links'' and could, for example, aggregate pros and cons of the different options as direct answers. However, changing the result presentation is an intricate decision such that the classification of comparative questions forms a highly precision-oriented task.

We introduce the use of Poincaré embeddings to improve existing state-of-the-art approaches to domain-specific taxonomy induction from text as a signal for both relocating wrong hyponym terms within a (pre-induced) taxonomy as well as for attaching disconnected terms in a taxonomy.This method substantially improves previous state-of-the-art results on the SemEval-2016 Task 13 on taxonomy extraction.We demonstrate the superiority of Poincaré embeddings over distributional semantic representations, supporting the hypothesis that they can better capture hierarchical lexical-semantic relationships than embeddings in the Euclidean space.

De-identification is the task of detecting protected health information (PHI) in medical text.It is a critical step in sanitizing electronic health records (EHRs) to be shared for research.Automatic de-identification classifiers can significantly speed up the sanitization process.However, obtaining a large and diverse dataset to train such a classifier that works well across many types of medical text poses a challenge as privacy laws prohibit the sharing of raw medical records.We introduce a method to create privacy-preserving shareable representations of medical text (i.e. they contain no PHI) that does not require expensive manual pseudonymization.These representations can be shared between organizations to create unified datasets for training de-identification models.Our representation allows training a simple LSTM-CRF de-identification model to an F 1 score of 97.4%, which is comparable to a strong baseline that exposes private information in its representation.A robust, widely available de-identification classifier based on our representation could potentially enable studies for which de-identification would otherwise be too costly.

The compositionality degree of multiword expressions indicates to what extent the meaning of a phrase can be derived from the meaning of its constituents and their grammatical relations. Prediction of (non)-compositionality is a task that has been frequently addressed with distributional semantic models. We introduce a novel technique to blend hierarchical information with distributional information for predicting compositionality. In particular, we use hypernymy information of the multiword and its constituents encoded in the form of the recently introduced Poincaré embeddings in addition to the distributional information to detect compositionality for noun phrases. Using a weighted average of the distributional similarity and a Poincaré similarity function, we obtain consistent and substantial, statistically significant improvement across three gold standard datasets over state-of-the-art models based on distributional information only. Unlike traditional approaches that solely use an unsupervised setting, we have also framed the problem as a supervised task, obtaining comparable improvements. Further, we publicly release our Poincaré embeddings, which are trained on the output of handcrafted lexical-syntactic patterns on a large corpus.

The computation of distance measures between nodes in graphs is inefficient and does not scale to large graphs.We explore dense vector representations as an effective way to approximate the same information: we introduce a simple yet efficient and effective approach for learning graph embeddings.Instead of directly operating on the graph structure, our method takes structural measures of pairwise node similarities into account and learns dense node representations reflecting user-defined graph distance measures, such as e.g. the shortest path distance or distance measures that take information beyond the graph structure into account.We demonstrate a speed-up of several orders of magnitude when predicting word similarity by vector operations on our embeddings as opposed to directly computing the respective path-based measures, while outperforming various other graph embeddings on semantic similarity and word sense disambiguation tasks and show evaluations on the WordNet graph and two knowledge base graphs.

We present CAM (comparative argumentative machine), a novel open-domain IR system to argumentatively compare objects with respect to information extracted from the Common Crawl. In a user study, the participants obtained 15% more accurate answers using CAM compared to a "traditional" keyword-based search and were 20% faster in finding the answer to comparative questions.

We present a detailed theoretical and computational analysis of the Watset meta-algorithm for fuzzy graph clustering, which has been found to be widely applicable in a variety of domains. This algorithm creates an intermediate representation of the input graph, which reflects the “ambiguity” of its nodes. Then, it uses hard clustering to discover clusters in this “disambiguated” intermediate graph. After outlining the approach and analyzing its computational complexity, we demonstrate that Watset shows competitive results in three applications: unsupervised synset induction from a synonymy graph, unsupervised semantic frame induction from dependency triples, and unsupervised semantic class induction from a distributional thesaurus. Our algorithm is generic and can also be applied to other networks of linguistic data.

Learning from a real-world data stream and continuously updating the model without explicit supervision is a new challenge for NLP applications with machine learning components. In this work, we have developed an adaptive learning system for text simplification, which improves the underlying learning-to-rank model from usage data, i.e. how users have employed the system for the task of simplification. Our experimental result shows that, over a period of time, the performance of the embedded paraphrase ranking model increases steadily improving from a score of 62.88% up to 75.70% based on the NDCG@10 evaluation metrics. To our knowledge, this is the first study where an NLP component is adaptively improved through usage.

We introduce "Unspeech" embeddings, which are based on unsupervised learning of context feature representations for spoken language. The embeddings were trained on up to 9500 hours of crawled English speech data without transcriptions or speaker information, by using a straightforward learning objective based on context and non-context discrimination with negative sampling. We use a Siamese convolutional neural network architecture to train Unspeech embeddings and evaluate them on speaker comparison, utterance clustering and as a context feature in TDNN-HMM acoustic models trained on TED-LIUM, comparing it to i-vector baselines. Particularly decoding out-of-domain speech data from the recently released Common Voice corpus shows consistent WER reductions. We release our source code and pre-trained Unspeech models under a permissive open source license.

Depending on downstream applications, it is advisable to extend the notion of tokenization from low-level character-based token boundary detection to identification of meaningful and useful language units. This entails both identifying units composed of several single words that form a several single words that form a, as well as splitting single-word compounds into their meaningful parts. In this article, we introduce unsupervised and knowledge-free methods for these two tasks. The main novelty of our research is based on the fact that methods are primarily based on distributional similarity, of which we use two flavors: a sparse count-based and a dense neural-based distributional semantic model. First, we introduce DRUID, which is a method for detecting MWEs. The evaluation on MWE-annotated data sets in two languages and newly extracted evaluation data sets for 32 languages shows that DRUID compares favorably over previous methods not utilizing distributional information. Second, we present SECOS, an algorithm for decompounding close compounds. In an evaluation of four dedicated decompounding data sets across four languages and on data sets extracted from Wiktionary for 14 languages, we demonstrate the superiority of our approach over unsupervised baselines, sometimes even matching the performance of previous language-specific and supervised methods. In a final experiment, we show how both decompounding and MWE information can be used in information retrieval. Here, we obtain the best results when combining word information with MWEs and the compound parts in a bag-of-words retrieval set-up. Overall, our methodology paves the way to automatic detection of lexical units beyond standard tokenization techniques without language-specific preprocessing steps such as POS tagging.

Abstract We present an approach to combining distributional semantic representations induced from text corpora with manually constructed lexical semantic networks. While both kinds of semantic resources are available with high lexical coverage, our aligned resource combines the domain specificity and availability of contextual information from distributional models with the conciseness and high quality of manually crafted lexical networks. We start with a distributional representation of induced senses of vocabulary terms, which are accompanied with rich context information given by related lexical items. We then automatically disambiguate such representations to obtain a full-fledged proto-conceptualization, i.e. a typed graph of induced word senses. In a final step, this proto-conceptualization is aligned to a lexical ontology, resulting in a hybrid aligned resource. Moreover, unmapped induced senses are associated with a semantic type in order to connect them to the core resource. Manual evaluations against ground-truth judgments for different stages of our method as well as an extrinsic evaluation on a knowledge-based Word Sense Disambiguation benchmark all indicate the high quality of the new hybrid resource. Additionally, we show the benefits of enriching top-down lexical knowledge resources with bottom-up distributional information from text for addressing high-end knowledge acquisition tasks such as cleaning hypernym graphs and learning taxonomies from scratch.

This paper presents a new graph-based approach that induces synsets using synonymy dictionaries and word embeddings.First, we build a weighted graph of synonyms extracted from commonly available resources, such as Wiktionary.Second, we apply word sense induction to deal with ambiguous words.Finally, we cluster the disambiguated version of the ambiguous input graph into synsets.Our meta-clustering approach lets us use an efficient hard clustering algorithm to perform a fuzzy clustering of the graph.Despite its simplicity, our approach shows excellent results, outperforming five competitive state-of-the-art methods in terms of F-score on three gold standard datasets for English and Russian derived from large-scale manually constructed lexical resources.

This paper presents Storyfinder, an application which consists of a browser plugin and a web server backend with the goal to highlight and manage the information contained in web pages by combining techniques from natural language processing and visual analytics. Webpages are analyzed while visiting them by means of natural language processing components, and metadata in the form of named entities and keywords are extracted and stored for further reference. The extracted information is instantaneously highlighted in the web page and stored in a graph of entities and relations. The graph can be inspected and modified. The investigational scope can be set to a single web page, multiple web pages, or the complete set of analyzed web pages in a user's history. The graph view is designed to adhere to standards of visual analytics and information visualization. Storyfinder is available as an open source application. Its benefit for information access is evaluated in a small user study.

Stefano Faralli, Alexander Panchenko, Chris Biemann, Simone Paolo Ponzetto. Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers. 2017.

Alexander Panchenko, Eugen Ruppert, Stefano Faralli, Simone Paolo Ponzetto, Chris Biemann. Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers. 2017.

We propose a framework for lexical substitution that is able to perform transfer learning across languages. Datasets for this task are available in at least three languages (English, Italian, and German). Previous work has addressed each of these tasks in isolation. In contrast, we regard the union of three shared tasks as a combined multilingual dataset. We show that a supervised system can be trained effectively, even if training and evaluation data are from different languages. Successful transfer learning between languages suggests that the learned model is in fact independent of the underlying language. We combine state-of-the-art unsupervised features obtained from syntactic word embeddings and distributional thesauri in a supervised delexicalized ranking system. Our system improves over state of the art in the full lexical substitution task in all three languages.

We present Ambient Search, an open source system for displaying and retrieving relevant documents in real time for speech input. The system works ambiently, that is, it unobstructively listens to speech streams in the background, identifies keywords and keyphrases for query construction and continuously serves relevant documents from its index. Query terms are ranked with Word2Vec and TF-IDF and are continuously updated to allow for ongoing querying of a document collection. The retrieved documents, in our case Wikipedia articles, are visualized in real time in a browser interface. Our evaluation shows that Ambient Search compares favorably to another implicit information retrieval system on speech streams. Furthermore, we extrinsically evaluate multiword keyphrase generation, showing positive impact for manual transcriptions.

In this paper we present a word decompounding method that is based on distributional semantics.Our method does not require any linguistic knowledge and is initialized using a large monolingual corpus.The core idea of our approach is that parts of compounds (like "candle" and "stick") are semantically similar to the entire compound, which helps to exclude spurious splits (like "candles" and "tick").We report results for German and Dutch: For German, our unsupervised method comes on par with the performance of a rule-based and a supervised method and significantly outperforms two unsupervised baselines.For Dutch, our method performs only slightly below a rule-based optimized compound splitter.

We present a new unsupervised mechanism, which ranks word n-grams according to their multiwordness.It heavily relies on a new uniqueness measure that computes, based on a distributional thesaurus, how often an n-gram could be replaced in context by a single-worded term.In addition with a downweighting mechanism for incomplete terms this forms a new measure called DRUID.Results show large improvements on two small test sets over competitive baselines.We demonstrate the scalability of the method to large corpora, and the independence of the measure of shallow syntactic filtering.

Omer Levy, Steffen Remus, Chris Biemann, Ido Dagan. Proceedings of the 2015 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2015.

A book aiming to build a bridge between two fields that share the subject of research but do not share the same views necessarily puts itself in a difficult position: The authors have either to strike a fair balance at peril of dissatisfying both sides or nail their colors to the mast and cater mainly to one of two communities. For semantic processing of natural language with either NLP methods or Semantic Web approaches, the authors clearly favor the latter and propose a strictly ontology-driven interpretation of natural language. The main contribution of the book, driving semantic processing from the ground up by a formal domain-specific ontology, is elaborated in ten well-structured chapters spanning 143 pages of content.Throughout the book, examples from the soccer domain excellently illustrate the main concepts in high machine-readable detail. The first chapter sets the scene with a motivating example: Humans have sufficient background knowledge to interpret a description of a soccer match and can grasp a lot of meaning beyond the literal content. For example, if team A scored two goals and team B eventually wins the match, team B scored at least three goals. To enable these and other kinds of inferences in machines, these need to be equipped with a domain ontology that formalizes domain knowledge and domain-specific reasoning, as well as with a mechanism to construct formal representations of natural language text. The key idea of this book is to place the ontology at the center of such an interpretation process: The domain and all its relevant semantic distinctions are defined in the ontology, thus the natural language semantic parser must only be aware of these. As opposed to generic tools such as Boxer (Bos 2008) that turn natural language into formal representations, driving the formalization of NL directly by its target ontology ensures that it can be directly consumed by further layers of interpretation, such as reasoners. The remainder of the first chapter gives a very short summary of the state of affairs in semantic interpretation and semantic parsing in NLP. Although this summary is sufficient to highlight what is missing for drawing inferences on top of natural language statements, it is—with a mere two pages—necessarily incomplete and omits even mainstream approaches such as frame semantic parsing or semantic role labeling. Further, the relation of the proposal to the Semantic Web is discussed, from which formats, interoperability, and description languages are leveraged. Primers on the RDF data format and on the soccer domain complete the chapter.Chapter 2 defines the concept of ontology. Great care is taken to discuss the definition of ontologies and various ontology description languages such as OWL 2 DL, as well as their expressivity. In Chapter 3, linguistic formalisms for representing syntax and semantics of natural language are discussed and ways to connect these to the ontology are introduced. Although, in principle, there are many possible formalisms to serve the “linguistic side,” the authors decide on Lexicalized Tree Adjoining Grammar (LTA, Schabes 1990) for syntactic processing. In the LTAG lexicon, each lexical entry links to all elementary trees (subtrees that specify the valid contexts in terms of constituents) it anchors, whereas the corresponding ontology grammar links each ontological concept to the projections of all lexical items that verbalize this concept. Thus, each concept is associated with an exhaustive enumeration of patterns it is expressed in. On the semantic level, Discourse Representation Theory (DRT; Kamp and Reyle 1993) is chosen as a formalism for semantic operations such as coreference and quantification. DRT and LTAG are subsequently paired in a representation called DUDES (Dependency-based Underspecified Discourse Representation structures; Cimiano 2009). Here, the DRT-inspired representations are again linked to ontology concepts. Alhough the choice of framework seems largely rooted in previous works of the first author, it is clearly stated that the connection to the ontology could also be realized for other syntactic (e.g., LFG, HPSG) and semantic (e.g., GLUE, MRS) frameworks.Chapter 4 deals with the representation of the ontological lexicons, which specify the interpretation of lexical items with respect to the target ontology. The declarative LEMON lexicon model for ontologies, following closely the approach to the lexicon of Ontological Semantics (Nirenburg and Raskin 2004), is laid out and exemplified in great detail. In the fifth chapter, the authors describe how the ontology grammar for DUDES—at least, the ontology-specific parts—can be generated from the lexicon. While this might seem merely a syntactic manipulation as the patterns of concept manifestation have been encoded in the lexicon, the point is to keep the lexicon separate from DUDES to allow for the generation of other types of ontology grammars. Note that different domain ontologies generate different grammars that could yield different interpretations for the same sentence. In Chapter 6, the overall interplay of previously discussed ingredients for semantic interpretation is exemplified, and challenges with respect to structural ambiguities and underspecification are pointed out. These are supposed to be left underspecified until they can be resolved with ontological reasoning, which is the subject of Chapter 7. Subsequently, the formalization of time in the framework is declared at length in Chapter 8, building on time interval calculi from the literature. In Chapter 9, the application of question answering over structured data is discussed—not to be confused with question answering from unstructured sources in the flavor of TREC or IBM Watson. In such a system, the natural language query is translated into a SPARQL query that can be run against public endpoints. While the examples successfully illustrate the translation from DUDES representations to SPARQL queries, no qualitative or quantitative evaluation is provided despite the existence of public benchmarks.The final chapter concludes with a very strong statement: The mainstream of computational linguistics research supposedly concentrates on domain-independent semantic representations, which is deemed “wrong,” as it would not do justice to domain distinctions and would not enable the connection to domain-specific ontologies to allow for domain-specific reasoning. Regretting that a principled use of ontologies in NLP is not widespread, the authors finally paint a vision of an ontology-based NLP ecosystem modeled after their approach. While proposing the use of statistical methods to overcome brittleness, lack of coverage, and the limitations of logical reasoning in the presence of uncertainty, it remains entirely unclear how this could be carried out.Throughout, the book contains exercises that greatly help in internalizing the discussed concepts. Also, an array of available resources including a demo of the question answering system is announced. Some of these, however, were not yet available on the companion Web site at the time of review.Even though the book states that its focus is “certainly not on robustness and coverage” (p. 6), I am still missing more concrete ideas on how robustness and coverage might be addressed than “by using machine learning techniques” (p. 142) without further elaboration. Thus, it is no coincidence that the implementation remains on the level of toy examples, however illustrating they might be. Also, while admitting to domain drift and the necessity of adaptation, it remains unclear how this adaptation, the integration of domain-specific and domain-independent processing, as well as the integration of several domains can be tractably attained. A deeper account on related approaches to computational semantics and a more recent bibliography on NLP/CL approaches could probably have avoided the impression that the interpretation of language as advocated here provides hardly more than other, same-old classic AI approaches.Also, the pledge to the field of NLP to use more of Semantic Web technologies and services is more Semantic Web–centric than I would have expected from a book “attempting to provide a step towards the synergy between these two fields” (p. xv): True, it is unfortunate that these two fields interact as little as they do, as both are concerned with understanding the semantics of text, but the authors' clearly voiced credo on how NLP should finally follow suit is not very helpful for creating such synergy. To bridge the gap, the Semantic Web community should probably start leveraging results and evaluation methodologies from CL/NLP instead of reinventing/ignoring them, and the NLP community probably has to understand that ontological grounding and resolving entities to URIs is in fact enabling applications that go well beyond dependency parsing and word sense disambiguation. The Semantic Web community should understand that statistical methods and unsupervised acquisition are not the enemy, but the key to scalable, domain-adaptive processing of natural language, which has been known to evade total formalization since Sapier's “all grammars leak.” The NLP community, especially statistical semantics, in turn, should understand the need and the demand for linking concepts to manually curated taxonomies and controlled vocabularies, as these are pervasive in industrial knowledge management. NLP should see the benefit in standardization for interoperability, and Semantic Webbers have to finally figure out that from format does not follow function.To sum up, a well-written book goes the extra mile to exemplify its core contribution, which is to consequently drive language processing by the ontology of a target application from the very start. The exercises, together with the online materials, make it a useful resource for teaching. Unlike a similar proposal by Nirenburg and Raskin (2004), it is constructed and exemplified with Semantic Web technology such as RDF, OWL, SPARQL, open data, standardization, reasoning, and domain ontology. For this reason, the book will supposedly be very well received in the Semantic Web community, but could be perceived controversially by the NLP community: While the idea to radically align natural language processing to a target-domain ontology is reasonable, well-motivated from the application point of view, and a worthwhile contribution, the book unfortunately does not succeed in convincing most modern computational linguists that this approach can overcome known limitations of previous similar rule-based and knowledge-driven approaches, such as brittleness, poor scalability, and little adaptivity.

Abstract In this paper, we propose an unsupervised and automated method to identify noun sense changes based on rigorous analysis of time-varying text data available in the form of millions of digitized books and millions of tweets posted per day. We construct distributional-thesauri-based networks from data at different time points and cluster each of them separately to obtain word-centric sense clusters corresponding to the different time points. Subsequently, we propose a split/join based approach to compare the sense clusters at two different time points to find if there is ‘birth’ of a new sense. The approach also helps us to find if an older sense was ‘split’ into more than one sense or a newer sense has been formed from the ‘join’ of older senses or a particular sense has undergone ‘death’. We use this completely unsupervised approach (a) within the Google books data to identify word sense differences within a media, and (b) across Google books and Twitter data to identify differences in word sense distribution across different media. We conduct a thorough evaluation of the proposed methodology both manually as well as through comparison with WordNet.

Sunny Mitra, Ritwik Mitra, Martin Riedl, Chris Biemann, Animesh Mukherjee, Pawan Goyal. Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2014.

Abstract Understanding relationships between people and organizations by reading newspaper articles is difficult to manage for humans due to the large amount of data. To address this problem, we present and evaluate a new visual analytics system, which offers interactive exploration and tagging of social networks extracted from newspapers. For the visual exploration of the network, we extract “interesting” neighbourhoods of nodes, using a new degree of interest (DOI) measure based on edges instead of nodes. It improves the seminal definition of DOI, which we find to produce the same “globally interesting” neighbourhoods in our use case, regardless of the query. Our approach allows answering different user queries appropriately, avoiding uniform search results. We propose a user‐driven pattern‐based classifier for discovery and tagging of non‐taxonomic semantic relations. Our approach does not require any a‐priori user knowledge, such as expertise in syntax or pattern creation. An evaluation shows that our classifier is capable of identifying known lexico‐syntactic patterns as well as various domain‐specific patters. Our classifier yields good results already with a small amount of training, and continuously improves through user feedback. We conduct a user study to evaluate whether our visual interactive system has an impact on how users tag relationships, as compared to traditional text‐based interfaces. Study results suggest that users of the visual system tend to tag more concisely, avoiding too abstract or overly specific relationship labels.

We introduce a new highly scalable approach for computing Distributional Thesauri (DTs).By employing pruning techniques and a distributed framework, we make the computation for very large corpora feasible on comparably small computational resources.We demonstrate this by releasing a DT for the whole vocabulary of Google Books syntactic n-grams.Evaluating against lexical resources using two measures, we show that our approach produces higher quality DTs than previous approaches, and is thus preferable in terms of speed and quality for large corpora.

Graphs are ubiquitous. There is hardly any domain in which objects and their relations cannot be intuitively represented as nodes and edges in a graph. Graph theory is a well-studied sub-discipline of mathematics, with a large body of results and a large number of efficient algorithms that operate on graphs. Like many other disciplines, the fields of natural language processing (NLP) and information retrieval (IR) also deal with data that can be represented as a graph. In this light, it is somewhat surprising that only in recent years the applicability of graph-theoretical frameworks to language technology became apparent and increasingly found its way into publications in the field of computational linguistics. Using algorithms that take the overall graph structure of a problem into account, rather than characteristics of single objects or (unstructured) sets of objects, graph-based methods have been shown to improve a wide range of NLP tasks. In a short but comprehensive overview of the field of graph-based methods for NLP and IR, Rada Mihalcea and Dragomir Radev list an extensive number of techniques and examples from a wide range of research papers by a large number of authors. This book provides an excellent review of this research area, and serves both as an introduction and as a survey of current graph-based techniques in NLP and IR. Because the few existing surveys in this field concentrate on particular aspects, such as graph clustering (Lancichinetti and Fortunato 2009) or IR (Liu 2006), a textbook on the topic was very much needed and this book surely fills this gap.The book is organized in four parts and contains a total of nine chapters. The first part gives an introduction to notions of graph theory, and the second part covers natural and random networks. The third part is devoted to graph-based IR, and part IV covers graph-based NLP. Chapter 1 lays the groundwork for the remainder of the book by introducing all necessary concepts in graph theory, including the notation, graph properties, and graph representations. In the second chapter, a glimpse is offered into the plethora of graph-based algorithms that have been developed independently of applications in NLP and IR. Sacrificing depth for breadth, this chapter does a great job in touching on a wide variety of methods, including minimum spanning trees, shortest-path algorithms, cuts and flows, subgraph matching, dimensionality reduction, random walks, spreading activation, and more. Algorithms are explained concisely, using examples, pseudo-code, and/or illustrations, some of which are very well suited for classroom examples. Network theory is presented in Chapter 3. The term network is here used to refer to naturally occurring relations, as opposed to graphs being generated by an automated process. After presenting the classical Erdős-Rényi random graph model and showing its inadequacy to model power-law degree distributions following Zipf's law, scale-free small-world networks are introduced. Further, several centrality measures, as well as other topics in network theory, are defined and exemplified.Establishing the connection to NLP, Chapter 4 introduces networks constructed from natural language. Co-occurrence networks and syntactic dependency networks are examined quantitatively. Results on the structure of semantic networks such as WordNet are presented, as well as a range of similarity networks between lexical units. This chapter will surely inspire the reader to watch out for networks in his/her own data. Chapter 5 turns to link analysis for the Web. The PageRank algorithm is described at length, variants for undirected and weighted graphs are introduced, and the algorithm's application to topic-sensitive analysis and query-dependent link analysis is discussed. This chapter is the only one that touches on core IR, and this is also the only chapter with content that can be found in other textbooks (e.g., Liu 2011). Still, this chapter is an important prerequisite for the chapter on applications. It would have been possible to move the description of the algorithms to Chapter 2, however, omitting this part.The topic of Chapter 6 is text clustering with graph-based methods, outlining the Fiedler method, the Kernighan–Lin method, min-cut clustering, betweenness, and random walk clustering. After defining measures on cluster quality for graphs, spectral and non-spectral graph clustering methods are briefly introduced. Most of the chapter is to be understood as a presentation of general graph clustering methods rather than their application to language. For this, some representative methods for different core ideas were selected. Part IV on graph-based NLP contains the chapters probably most interesting to readers working in computational linguistics. In Chapter 7, graph-based methods for lexical semantics are presented, including detection of semantic classes, synonym detection using random walks on semantic networks, semantic distance on WordNet, and textual entailment using graph matching. Methods for word sense and name disambiguation with graph clustering and random walks are described. The chapter closes with graph-based methods for sentiment lexicon construction and subjectivity classification.Graph-based methods for syntactic processing are presented in Chapter 8: an unsupervised part-of-speech tagging algorithm based on graph clustering, minimum spanning trees for dependency parsing, PP-attachment with random walks over syntactic co-occurrence graphs, and coreference resolution with graph cuts. In the final chapter, many of the algorithms introduced in the previous chapters are applied to NLP applications as diverse as summarization, passage retrieval, keyword extraction, topic identification and segmentation, discourse, machine translation, cross-language IR, term weighting, and question answering.As someone with a background in graph-based NLP, I enjoyed reading this book. The writing style is concise and clear, and the authors succeed in conveying the most important points from an incredibly large number of works, viewed from the graph-based perspective. I also liked the extensive use of examples—throughout, almost half of the space is used for figures and tables illustrating the methods, which some readers might perceive as unbalanced, however. With just under 200 pages and a topic as broad as this, it necessarily follows that many of the presented methods are exemplified and touched upon rather than discussed in great detail. Although this sometimes leads to the situation that some passages can only be understood with background knowledge, it is noteworthy that every chapter includes a section on further reading. In this way, the book serves as an entry point to a deeper engagement with graph-based methods for NLP and IR, and it encourages readers to see their NLP problem from a graph-based view.For a future edition, however, I have a few wishes: It would be nice if the figures and examples were less detached from the text and explained more thoroughly. At times, it would be helpful to present deeper insights and to connect the methodologies, rather than just presenting them next to each other. Also, some of the definitions in Chapter 2 could be less confusing and structured better.Because this book emphasizes graph-based aspects for language processing rather than aiming at exhaustively treating the numerous tasks that benefit from graph-based methods, it cannot replace a general introduction to NLP or IR: For students without prior knowledge in NLP and IR, a more guided and focused approach to the topic would be required. The target audience is, rather, NLP researchers and professionals who want to add the graph-based view to their arsenal of methods, and to become inspired by this rapidly growing research area. It is equally suited for people working in graph algorithms to learn about graphs in language as a field of application for their work. I will surely consult this volume in the future to supplement the preparation of lectures because of its comprehensive references and its richness in examples.