Leon Derczynski

Leon Derczynski, Hannah Rose Kirk, Vidhisha Balachandran et al. · cmu, oxford

This paper introduces RiskCards, a framework for structured assessment and documentation of risks associated with an application of language models. As with all language, text generated by language models can be harmful, or used to bring about harm. Automating language generation adds both an element of scale and also more subtle or emergent undesirable tendencies to the generated text. Prior work establishes a wide variety of language model harms to many different actors: existing taxonomies identify categories of harms posed by language models; benchmarks establish automated tests of these harms; and documentation standards for models, tasks and datasets encourage transparent reporting. However, there is no risk-centric framework for documenting the complexity of a landscape in which some risks are shared across models and contexts, while others are specific, and where certain conditions may be required for risks to manifest as harms. RiskCards address this methodological gap by providing a generic framework for assessing the use of a given language model in a given scenario. Each RiskCard makes clear the routes for the risk to manifest harm, their placement in harm taxonomies, and example prompt-output pairs. While RiskCards are designed to be open-source, dynamic and participatory, we present a "starter set" of RiskCards taken from a broad literature survey, each of which details a concrete risk presentation. Language model RiskCards initiate a community knowledge base which permits the mapping of risks and harms to a specific model or its application scenario, ultimately contributing to a better, safer and shared understanding of the risk landscape.

Aaron Blakeman, Aarti Basant, Abhinav Khattar et al. · nvidia

As inference-time scaling becomes critical for enhanced reasoning capabilities, it is increasingly becoming important to build models that are efficient to infer. We introduce Nemotron-H, a family of 8B and 56B/47B hybrid Mamba-Transformer models designed to reduce inference cost for a given accuracy level. To achieve this goal, we replace the majority of self-attention layers in the common Transformer model architecture with Mamba layers that perform constant computation and require constant memory per generated token. We show that Nemotron-H models offer either better or on-par accuracy compared to other similarly-sized state-of-the-art open-sourced Transformer models (e.g., Qwen-2.5-7B/72B and Llama-3.1-8B/70B), while being up to 3$\times$ faster at inference. To further increase inference speed and reduce the memory required at inference time, we created Nemotron-H-47B-Base from the 56B model using a new compression via pruning and distillation technique called MiniPuzzle. Nemotron-H-47B-Base achieves similar accuracy to the 56B model, but is 20% faster to infer. In addition, we introduce an FP8-based training recipe and show that it can achieve on par results with BF16-based training. This recipe is used to train the 56B model. We are releasing Nemotron-H base model checkpoints with support in Hugging Face and NeMo.

Aakshita Chandiramani, Aaron Blakeman, Abdullahi Olaoye et al. · amazon-science, cmu

We describe the pre-training, post-training, and quantization of Nemotron 3 Super, a 120 billion (active 12 billion) parameter hybrid Mamba-Attention Mixture-of-Experts model. Nemotron 3 Super is the first model in the Nemotron 3 family to 1) be pre-trained in NVFP4, 2) leverage LatentMoE, a new Mixture-of-Experts architecture that optimizes for both accuracy per FLOP and accuracy per parameter, and 3) include MTP layers for inference acceleration through native speculative decoding. We pre-trained Nemotron 3 Super on 25 trillion tokens followed by post-training using supervised fine tuning (SFT) and reinforcement learning (RL). The final model supports up to 1M context length and achieves comparable accuracy on common benchmarks, while also achieving up to 2.2x and 7.5x higher inference throughput compared to GPT-OSS-120B and Qwen3.5-122B, respectively. Nemotron 3 Super datasets, along with the base, post-trained, and quantized checkpoints, are open-sourced on HuggingFace.

Marcos Treviso, Ji-Ung Lee, Tianchu Ji et al. · uw

Recent work in natural language processing (NLP) has yielded appealing results from scaling model parameters and training data; however, using only scale to improve performance means that resource consumption also grows. Such resources include data, time, storage, or energy, all of which are naturally limited and unevenly distributed. This motivates research into efficient methods that require fewer resources to achieve similar results. This survey synthesizes and relates current methods and findings in efficient NLP. We aim to provide both guidance for conducting NLP under limited resources, and point towards promising research directions for developing more efficient methods.

Aarti Basant, Abhijit Khairnar, Abhijit Paithankar et al. · nvidia

We introduce Nemotron-Nano-9B-v2, a hybrid Mamba-Transformer language model designed to increase throughput for reasoning workloads while achieving state-of-the-art accuracy compared to similarly-sized models. Nemotron-Nano-9B-v2 builds on the Nemotron-H architecture, in which the majority of the self-attention layers in the common Transformer architecture are replaced with Mamba-2 layers, to achieve improved inference speed when generating the long thinking traces needed for reasoning. We create Nemotron-Nano-9B-v2 by first pre-training a 12-billion-parameter model (Nemotron-Nano-12B-v2-Base) on 20 trillion tokens using an FP8 training recipe. After aligning Nemotron-Nano-12B-v2-Base, we employ the Minitron strategy to compress and distill the model with the goal of enabling inference on up to 128k tokens on a single NVIDIA A10G GPU (22GiB of memory, bfloat16 precision). Compared to existing similarly-sized models (e.g., Qwen3-8B), we show that Nemotron-Nano-9B-v2 achieves on-par or better accuracy on reasoning benchmarks while achieving up to 6x higher inference throughput in reasoning settings like 8k input and 16k output tokens. We are releasing Nemotron-Nano-9B-v2, Nemotron-Nano12B-v2-Base, and Nemotron-Nano-9B-v2-Base checkpoints along with the majority of our pre- and post-training datasets on Hugging Face.

Hannah Rose Kirk, Abeba Birhane, Bertie Vidgen et al. · oxford

Text data can pose a risk of harm. However, the risks are not fully understood, and how to handle, present, and discuss harmful text in a safe way remains an unresolved issue in the NLP community. We provide an analytical framework categorising harms on three axes: (1) the harm type (e.g., misinformation, hate speech or racial stereotypes); (2) whether a harm is \textit{sought} as a feature of the research design if explicitly studying harmful content (e.g., training a hate speech classifier), versus \textit{unsought} if harmful content is encountered when working on unrelated problems (e.g., language generation or part-of-speech tagging); and (3) who it affects, from people (mis)represented in the data to those handling the data and those publishing on the data. We provide advice for practitioners, with concrete steps for mitigating harm in research and in publication. To assist implementation we introduce \textsc{HarmCheck} -- a documentation standard for handling and presenting harmful text in research.

Ji-Ung Lee, Haritz Puerto, Betty van Aken et al. · allen-ai, cmu

Many recent improvements in NLP stem from the development and use of large pre-trained language models (PLMs) with billions of parameters. Large model sizes makes computational cost one of the main limiting factors for training and evaluating such models; and has raised severe concerns about the sustainability, reproducibility, and inclusiveness for researching PLMs. These concerns are often based on personal experiences and observations. However, there had not been any large-scale surveys that investigate them. In this work, we provide a first attempt to quantify these concerns regarding three topics, namely, environmental impact, equity, and impact on peer reviewing. By conducting a survey with 312 participants from the NLP community, we capture existing (dis)parities between different and within groups with respect to seniority, academia, and industry; and their impact on the peer reviewing process. For each topic, we provide an analysis and devise recommendations to mitigate found disparities, some of which already successfully implemented. Finally, we discuss additional concerns raised by many participants in free-text responses.

Amala Sanjay Deshmukh, Kateryna Chumachenko, Tuomas Rintamaki et al. · amazon-science, nvidia

We introduce Nemotron 3 Nano Omni, the latest model in the Nemotron multimodal series and the first to natively support audio inputs alongside text, images, and video. Nemotron 3 Nano Omni delivers consistent accuracy improvements over its predecessor, Nemotron Nano V2 VL, across all modalities, enabled by advances in architecture, training data and recipes. In particular, Nemotron 3 delivers leading results in real-world document understanding, long audio-video comprehension, and agentic computer use. Built on the highly efficient Nemotron 3 Nano 30B-A3B backbone, Nemotron 3 Nano Omni further incorporates innovative multimodal token-reduction techniques to deliver substantially lower inference latency and higher throughput than other models of similar size. We are releasing model checkpoints in BF16, FP8, and FP4 formats, along with portions of the training data and codebase to facilitate further research and development.

Nanna Inie, Jonathan Stray, Leon Derczynski · uw

Engaging in the deliberate generation of abnormal outputs from Large Language Models (LLMs) by attacking them is a novel human activity. This paper presents a thorough exposition of how and why people perform such attacks, defining LLM red-teaming based on extensive and diverse evidence. Using a formal qualitative methodology, we interviewed dozens of practitioners from a broad range of backgrounds, all contributors to this novel work of attempting to cause LLMs to fail. We focused on the research questions of defining LLM red teaming, uncovering the motivations and goals for performing the activity, and characterizing the strategies people use when attacking LLMs. Based on the data, LLM red teaming is defined as a limit-seeking, non-malicious, manual activity, which depends highly on a team-effort and an alchemist mindset. It is highly intrinsically motivated by curiosity, fun, and to some degrees by concerns for various harms of deploying LLMs. We identify a taxonomy of 12 strategies and 35 different techniques of attacking LLMs. These findings are presented as a comprehensive grounded theory of how and why people attack large language models: LLM red teaming.

Aaron Blakeman, Aaron Grattafiori, Aarti Basant et al. · nvidia

We present Nemotron 3 Nano 30B-A3B, a Mixture-of-Experts hybrid Mamba-Transformer language model. Nemotron 3 Nano was pretrained on 25 trillion text tokens, including more than 3 trillion new unique tokens over Nemotron 2, followed by supervised fine tuning and large-scale RL on diverse environments. Nemotron 3 Nano achieves better accuracy than our previous generation Nemotron 2 Nano while activating less than half of the parameters per forward pass. It achieves up to 3.3x higher inference throughput than similarly-sized open models like GPT-OSS-20B and Qwen3-30B-A3B-Thinking-2507, while also being more accurate on popular benchmarks. Nemotron 3 Nano demonstrates enhanced agentic, reasoning, and chat abilities and supports context lengths up to 1M tokens. We release both our pretrained Nemotron 3 Nano 30B-A3B Base and post-trained Nemotron 3 Nano 30B-A3B checkpoints on Hugging Face.

Aaron Blakeman, Aaron Grattafiori, Aarti Basant et al. · nvidia

We introduce the Nemotron 3 family of models - Nano, Super, and Ultra. These models deliver strong agentic, reasoning, and conversational capabilities. The Nemotron 3 family uses a Mixture-of-Experts hybrid Mamba-Transformer architecture to provide best-in-class throughput and context lengths of up to 1M tokens. Super and Ultra models are trained with NVFP4 and incorporate LatentMoE, a novel approach that improves model quality. The two larger models also include MTP layers for faster text generation. All Nemotron 3 models are post-trained using multi-environment reinforcement learning enabling reasoning, multi-step tool use, and support granular reasoning budget control. Nano, the smallest model, outperforms comparable models in accuracy while remaining extremely cost-efficient for inference. Super is optimized for collaborative agents and high-volume workloads such as IT ticket automation. Ultra, the largest model, provides state-of-the-art accuracy and reasoning performance. Nano is released together with its technical report and this white paper, while Super and Ultra will follow in the coming months. We will openly release the model weights, pre- and post-training software, recipes, and all data for which we hold redistribution rights.

Manuel R. Ciosici, Leon Derczynski

Training large neural language models on large datasets is resource- and time-intensive. These requirements create a barrier to entry, where those with fewer resources cannot build competitive models. This paper presents various techniques for making it possible to (a) train a large language model using resources that a modest research lab might have, and (b) train it in a reasonable amount of time. We provide concrete recommendations for practitioners, which we illustrate with a case study: a T5 model for Danish, the first for this language.

Gcinizwe Dlamini, Imad Eddine Ibrahim Bekkouch, Adil Khan et al.

Misinformation has become a major concern in recent last years given its spread across our information sources. In the past years, many NLP tasks have been introduced in this area, with some systems reaching good results on English language datasets. Existing AI based approaches for fighting misinformation in literature suggest automatic stance detection as an integral first step to success. Our paper aims at utilizing this progress made for English to transfers that knowledge into other languages, which is a non-trivial task due to the domain gap between English and the target languages. We propose a black-box non-intrusive method that utilizes techniques from Domain Adaptation to reduce the domain gap, without requiring any human expertise in the target language, by leveraging low-quality data in both a supervised and unsupervised manner. This allows us to rapidly achieve similar results for stance detection for the Zulu language, the target language in this work, as are found for English. We also provide a stance detection dataset in the Zulu language. Our experimental results show that by leveraging English datasets and machine translation we can increase performances on both English data along with other languages.

Mateusz Jurewicz, Leon Derczynski

The task of learning to map an input set onto a permuted sequence of its elements is challenging for neural networks. Set-to-sequence problems occur in natural language processing, computer vision and structure prediction, where interactions between elements of large sets define the optimal output. Models must exhibit relational reasoning, handle varying cardinalities and manage combinatorial complexity. Previous attention-based methods require $n$ layers of their set transformations to explicitly represent $n$-th order relations. Our aim is to enhance their ability to efficiently model higher-order interactions through an additional interdependence component. We propose a novel neural set encoding method called the Set Interdependence Transformer, capable of relating the set's permutation invariant representation to its elements within sets of any cardinality. We combine it with a permutation learning module into a complete, 3-part set-to-sequence model and demonstrate its state-of-the-art performance on a number of tasks. These range from combinatorial optimization problems, through permutation learning challenges on both synthetic and established NLP datasets for sentence ordering, to a novel domain of product catalog structure prediction. Additionally, the network's ability to generalize to unseen sequence lengths is investigated and a comparative empirical analysis of the existing methods' ability to learn higher-order interactions is provided.

Jeppe Nørregaard, Leon Derczynski

Measuring inter-annotator agreement is important for annotation tasks, but many metrics require a fully-annotated set of data, where all annotators annotate all samples. We define Sparse Probability of Agreement, SPA, which estimates the probability of agreement when not all annotator-item-pairs are available. We show that under certain conditions, SPA is an unbiased estimator, and we provide multiple weighing schemes for handling data with various degrees of annotation.

Leon Derczynski, Annika Solveig Hedegaard Isfeldt, Signhild Djurhuus

This article documents a dataset of sentence pairs between Faroese and Danish, produced at ITU Copenhagen. The data covers tranlsation from both source languages, and is intended for use as training data for machine translation systems in this language pair.

Akhiad Bercovich, Itay Levy, Izik Golan et al. · nvidia

We introduce the Llama-Nemotron series of models, an open family of heterogeneous reasoning models that deliver exceptional reasoning capabilities, inference efficiency, and an open license for enterprise use. The family comes in three sizes -- Nano (8B), Super (49B), and Ultra (253B) -- and performs competitively with state-of-the-art reasoning models such as DeepSeek-R1 while offering superior inference throughput and memory efficiency. In this report, we discuss the training procedure for these models, which entails using neural architecture search from Llama 3 models for accelerated inference, knowledge distillation, and continued pretraining, followed by a reasoning-focused post-training stage consisting of two main parts: supervised fine-tuning and large scale reinforcement learning. Llama-Nemotron models are the first open-source models to support a dynamic reasoning toggle, allowing users to switch between standard chat and reasoning modes during inference. To further support open research and facilitate model development, we provide the following resources: 1. We release the Llama-Nemotron reasoning models -- LN-Nano, LN-Super, and LN-Ultra -- under the commercially permissive NVIDIA Open Model License Agreement. 2. We release the complete post-training dataset: Llama-Nemotron-Post-Training-Dataset. 3. We also release our training codebases: NeMo, NeMo-Aligner, and Megatron-LM.

Leon Derczynski, Robert Gaizauskas

We present CAVaT, a tool that performs Corpus Analysis and Validation for TimeML. CAVaT is an open source, modular checking utility for statistical analysis of features specific to temporally-annotated natural language corpora. It provides reporting, highlights salient links between a variety of general and time-specific linguistic features, and also validates a temporal annotation to ensure that it is logically consistent and sufficiently annotated. Uniquely, CAVaT provides analysis specific to TimeML-annotated temporal information. TimeML is a standard for annotating temporal information in natural language text. In this paper, we present the reporting part of CAVaT, and then its error-checking ability, including the workings of several novel TimeML document verification methods. This is followed by the execution of some example tasks using the tool to show relations between times, events, signals and links. We also demonstrate inconsistencies in a TimeML corpus (TimeBank) that have been detected with CAVaT.

Bertie Vidgen, Adarsh Agrawal, Ahmed M. Ahmed et al. · deepmind, oxford

This paper introduces v0.5 of the AI Safety Benchmark, which has been created by the MLCommons AI Safety Working Group. The AI Safety Benchmark has been designed to assess the safety risks of AI systems that use chat-tuned language models. We introduce a principled approach to specifying and constructing the benchmark, which for v0.5 covers only a single use case (an adult chatting to a general-purpose assistant in English), and a limited set of personas (i.e., typical users, malicious users, and vulnerable users). We created a new taxonomy of 13 hazard categories, of which 7 have tests in the v0.5 benchmark. We plan to release version 1.0 of the AI Safety Benchmark by the end of 2024. The v1.0 benchmark will provide meaningful insights into the safety of AI systems. However, the v0.5 benchmark should not be used to assess the safety of AI systems. We have sought to fully document the limitations, flaws, and challenges of v0.5. This release of v0.5 of the AI Safety Benchmark includes (1) a principled approach to specifying and constructing the benchmark, which comprises use cases, types of systems under test (SUTs), language and context, personas, tests, and test items; (2) a taxonomy of 13 hazard categories with definitions and subcategories; (3) tests for seven of the hazard categories, each comprising a unique set of test items, i.e., prompts. There are 43,090 test items in total, which we created with templates; (4) a grading system for AI systems against the benchmark; (5) an openly available platform, and downloadable tool, called ModelBench that can be used to evaluate the safety of AI systems on the benchmark; (6) an example evaluation report which benchmarks the performance of over a dozen openly available chat-tuned language models; (7) a test specification for the benchmark.

Bo Adler, Niket Agarwal, Ashwath Aithal et al. · nvidia

We release the Nemotron-4 340B model family, including Nemotron-4-340B-Base, Nemotron-4-340B-Instruct, and Nemotron-4-340B-Reward. Our models are open access under the NVIDIA Open Model License Agreement, a permissive model license that allows distribution, modification, and use of the models and its outputs. These models perform competitively to open access models on a wide range of evaluation benchmarks, and were sized to fit on a single DGX H100 with 8 GPUs when deployed in FP8 precision. We believe that the community can benefit from these models in various research studies and commercial applications, especially for generating synthetic data to train smaller language models. Notably, over 98% of data used in our model alignment process is synthetically generated, showcasing the effectiveness of these models in generating synthetic data. To further support open research and facilitate model development, we are also open-sourcing the synthetic data generation pipeline used in our model alignment process.

Aishwarya Padmakumar, Leon Derczynski, Traian Rebedea et al.

Automated methods for red teaming LLMs are an important tool to identify LLM vulnerabilities that may not be covered in static benchmarks, allowing for more thorough probing. They can also adapt to each specific LLM to discover weaknesses unique to it. Most current automated red teaming methods are intended for tackling safety and content moderation. Thus, they make use of content safety models as evaluators and optimize for circumventing them, and as such, have not been tested with other adversarial intents not typically captured by these. We propose a pipeline for training a red teaming model that can generalize to arbitrary adversarial goals, including objectives it has not been directly trained on, and that does not depend on the existence of a pre-existing evaluator available at training time. We demonstrate that finetuning small models, such as Qwen3-8B, using this pipeline results in a substantial improvement in their ability to generate attacks for both in and out of domain adversarial goals.

Arjun Krishna, Erick Galinkin, Leon Derczynski et al.

Large Language Models (LLMs) have become an essential tool in the programmer's toolkit, but their tendency to hallucinate code can be used by malicious actors to introduce vulnerabilities to broad swathes of the software supply chain. In this work, we analyze package hallucination behaviour in LLMs across popular programming languages examining both existing package references and fictional dependencies. By analyzing this package hallucination behaviour we find potential attacks and suggest defensive strategies to defend against these attacks. We discover that package hallucination rate is predicated not only on model choice, but also programming language, model size, and specificity of the coding task request. The Pareto optimality boundary between code generation performance and package hallucination is sparsely populated, suggesting that coding models are not being optimized for secure code. Additionally, we find an inverse correlation between package hallucination rate and the HumanEval coding benchmark, offering a heuristic for evaluating the propensity of a model to hallucinate packages. Our metrics, findings and analyses provide a base for future models, securing AI-assisted software development workflows against package supply chain attacks.

Heather Lent, Erick Galinkin, Yiyi Chen et al.

As NLP models are used by a growing number of end-users, an area of increasing importance is NLP Security (NLPSec): assessing the vulnerability of models to malicious attacks and developing comprehensive countermeasures against them. While work at the intersection of NLP and cybersecurity has the potential to create safer NLP for all, accidental oversights can result in tangible harm (e.g., breaches of privacy or proliferation of malicious models). In this emerging field, however, the research ethics of NLP have not yet faced many of the long-standing conundrums pertinent to cybersecurity, until now. We thus examine contemporary works across NLPSec, and explore their engagement with cybersecurity's ethical norms. We identify trends across the literature, ultimately finding alarming gaps on topics like harm minimization and responsible disclosure. To alleviate these concerns, we provide concrete recommendations to help NLP researchers navigate this space more ethically, bridging the gap between traditional cybersecurity and NLP ethics, which we frame as ``white hat NLP''. The goal of this work is to help cultivate an intentional culture of ethical research for those working in NLP Security.

Leon Derczynski, Erick Galinkin, Jeffrey Martin et al.

As Large Language Models (LLMs) are deployed and integrated into thousands of applications, the need for scalable evaluation of how models respond to adversarial attacks grows rapidly. However, LLM security is a moving target: models produce unpredictable output, are constantly updated, and the potential adversary is highly diverse: anyone with access to the internet and a decent command of natural language. Further, what constitutes a security weak in one context may not be an issue in a different context; one-fits-all guardrails remain theoretical. In this paper, we argue that it is time to rethink what constitutes ``LLM security'', and pursue a holistic approach to LLM security evaluation, where exploration and discovery of issues are central. To this end, this paper introduces garak (Generative AI Red-teaming and Assessment Kit), a framework which can be used to discover and identify vulnerabilities in a target LLM or dialog system. garak probes an LLM in a structured fashion to discover potential vulnerabilities. The outputs of the framework describe a target model's weaknesses, contribute to an informed discussion of what composes vulnerabilities in unique contexts, and can inform alignment and policy discussions for LLM deployment.

Erida Nurce, Jorgel Keci, Leon Derczynski

The ever growing usage of social media in the recent years has had a direct impact on the increased presence of hate speech and offensive speech in online platforms. Research on effective detection of such content has mainly focused on English and a few other widespread languages, while the leftover majority fail to have the same work put into them and thus cannot benefit from the steady advancements made in the field. In this paper we present \textsc{Shaj}, an annotated Albanian dataset for hate speech and offensive speech that has been constructed from user-generated content on various social media platforms. Its annotation follows the hierarchical schema introduced in OffensEval. The dataset is tested using three different classification models, the best of which achieves an F1 score of 0.77 for the identification of offensive language, 0.64 F1 score for the automatic categorization of offensive types and lastly, 0.52 F1 score for the offensive language target identification.

Magnus Jacobsen, Mikkel H. Sørensen, Leon Derczynski

Improvement in machine learning-based NLP performance are often presented with bigger models and more complex code. This presents a trade-off: better scores come at the cost of larger tools; bigger models tend to require more during training and inference time. We present multiple methods for measuring the size of a model, and for comparing this with the model's performance. In a case study over part-of-speech tagging, we then apply these techniques to taggers for eight languages and present a novel analysis identifying which taggers are size-performance optimal. Results indicate that some classical taggers place on the size-performance skyline across languages. Further, although the deep models have highest performance for multiple scores, it is often not the most complex of these that reach peak performance.

René Haas, Leon Derczynski

Automatic language identification is a challenging problem. Discriminating between closely related languages is especially difficult. This paper presents a machine learning approach for automatic language identification for the Nordic languages, which often suffer miscategorisation by existing state-of-the-art tools. Concretely we will focus on discrimination between six Nordic languages: Danish, Swedish, Norwegian (Nynorsk), Norwegian (Bokmål), Faroese and Icelandic.

Leon Derczynski

The power that machine learning models consume when making predictions can be affected by a model's architecture. This paper presents various estimates of power consumption for a range of different activation functions, a core factor in neural network model architecture design. Substantial differences in hardware performance exist between activation functions. This difference informs how power consumption in machine learning models can be reduced.

Marcos Zampieri, Preslav Nakov, Sara Rosenthal et al.

We present the results and main findings of SemEval-2020 Task 12 on Multilingual Offensive Language Identification in Social Media (OffensEval 2020). The task involves three subtasks corresponding to the hierarchical taxonomy of the OLID schema (Zampieri et al., 2019a) from OffensEval 2019. The task featured five languages: English, Arabic, Danish, Greek, and Turkish for Subtask A. In addition, English also featured Subtasks B and C. OffensEval 2020 was one of the most popular tasks at SemEval-2020 attracting a large number of participants across all subtasks and also across all languages. A total of 528 teams signed up to participate in the task, 145 teams submitted systems during the evaluation period, and 70 submitted system description papers.

Bertie Vidgen, Leon Derczynski

Data-driven analysis and detection of abusive online content covers many different tasks, phenomena, contexts, and methodologies. This paper systematically reviews abusive language dataset creation and content in conjunction with an open website for cataloguing abusive language data. This collection of knowledge leads to a synthesis providing evidence-based recommendations for practitioners working with this complex and highly diverse data.

Nanna Inie, Jeanette Falk Olesen, Leon Derczynski

Misinformation spread presents a technological and social threat to society. With the advance of AI-based language models, automatically generated texts have become difficult to identify and easy to create at scale. We present "The Rumour Mill", a playful art piece, designed as a commentary on the spread of rumours and automatically-generated misinformation. The mill is a tabletop interactive machine, which invites a user to experience the process of creating believable text by interacting with different tangible controls on the mill. The user manipulates visible parameters to adjust the genre and type of an automatically generated text rumour. The Rumour Mill is a physical demonstration of the state of current technology and its ability to generate and manipulate natural language text, and of the act of starting and spreading rumours.

Gudbjartur Ingi Sigurbergsson, Leon Derczynski

The presence of offensive language on social media platforms and the implications this poses is becoming a major concern in modern society. Given the enormous amount of content created every day, automatic methods are required to detect and deal with this type of content. Until now, most of the research has focused on solving the problem for the English language, while the problem is multilingual. We construct a Danish dataset containing user-generated comments from \textit{Reddit} and \textit{Facebook}. It contains user generated comments from various social media platforms, and to our knowledge, it is the first of its kind. Our dataset is annotated to capture various types and target of offensive language. We develop four automatic classification systems, each designed to work for both the English and the Danish language. In the detection of offensive language in English, the best performing system achieves a macro averaged F1-score of $0.74$, and the best performing system for Danish achieves a macro averaged F1-score of $0.70$. In the detection of whether or not an offensive post is targeted, the best performing system for English achieves a macro averaged F1-score of $0.62$, while the best performing system for Danish achieves a macro averaged F1-score of $0.73$. Finally, in the detection of the target type in a targeted offensive post, the best performing system for English achieves a macro averaged F1-score of $0.56$, and the best performing system for Danish achieves a macro averaged F1-score of $0.63$. Our work for both the English and the Danish language captures the type and targets of offensive language, and present automatic methods for detecting different kinds of offensive language such as hate speech and cyberbullying.

Leon Derczynski

This technical note describes a set of baseline tools for automatic processing of Danish text. The tools are machine-learning based, using natural language processing models trained over previously annotated documents. They are maintained at ITU Copenhagen and will always be freely available.

Genevieve Gorrell, Kalina Bontcheva, Leon Derczynski et al.

This is the proposal for RumourEval-2019, which will run in early 2019 as part of that year's SemEval event. Since the first RumourEval shared task in 2017, interest in automated claim validation has greatly increased, as the dangers of "fake news" have become a mainstream concern. Yet automated support for rumour checking remains in its infancy. For this reason, it is important that a shared task in this area continues to provide a focus for effort, which is likely to increase. We therefore propose a continuation in which the veracity of further rumours is determined, and as previously, supportive of this goal, tweets discussing them are classified according to the stance they take regarding the rumour. Scope is extended compared with the first RumourEval, in that the dataset is substantially expanded to include Reddit as well as Twitter data, and additional languages are also included.

Nikita Lozhnikov, Leon Derczynski, Manuel Mazzara

Stance detection is a critical component of rumour and fake news identification. It involves the extraction of the stance a particular author takes related to a given claim, both expressed in text. This paper investigates stance classification for Russian. It introduces a new dataset, RuStance, of Russian tweets and news comments from multiple sources, covering multiple stories, as well as text classification approaches to stance detection as benchmarks over this data in this language. As well as presenting this openly-available dataset, the first of its kind for Russian, the paper presents a baseline for stance prediction in the language.

Leon Derczynski, Kenny Meesters, Kalina Bontcheva et al.

Crisis responders are increasingly using social media, data and other digital sources of information to build a situational understanding of a crisis situation in order to design an effective response. However with the increased availability of such data, the challenge of identifying relevant information from it also increases. This paper presents a successful automatic approach to handling this problem. Messages are filtered for informativeness based on a definition of the concept drawn from prior research and crisis response experts. Informative messages are tagged for actionable data -- for example, people in need, threats to rescue efforts, changes in environment, and so on. In all, eight categories of actionability are identified. The two components -- informativeness and actionability classification -- are packaged together as an openly-available tool called Emina (Emergent Informativeness and Actionability).

Leon Derczynski, Matthew Rowe

Existing studies of how information diffuses across social networks have thus far concentrated on analysing and recovering the spread of deterministic innovations such as URLs, hashtags, and group membership. However investigating how mentions of real-world entities appear and spread has yet to be explored, largely due to the computationally intractable nature of performing large-scale entity extraction. In this paper we present, to the best of our knowledge, one of the first pieces of work to closely examine the diffusion of named entities on social media, using Reddit as our case study platform. We first investigate how named entities can be accurately recognised and extracted from discussion posts. We then use these extracted entities to study the patterns of entity cascades and how the probability of a user adopting an entity (i.e. mentioning it) is associated with exposures to the entity. We put these pieces together by presenting a parallelised diffusion model that can forecast the probability of entity adoption, finding that the influence of adoption between users can be characterised by their prior interactions -- as opposed to whether the users propagated entity-adoptions beforehand. Our findings have important implications for researchers studying influence and language, and for community analysts who wish to understand entity-level influence dynamics.

Ahmet Aker, Leon Derczynski, Kalina Bontcheva

Stance classification determines the attitude, or stance, in a (typically short) text. The task has powerful applications, such as the detection of fake news or the automatic extraction of attitudes toward entities or events in the media. This paper describes a surprisingly simple and efficient classification approach to open stance classification in Twitter, for rumour and veracity classification. The approach profits from a novel set of automatically identifiable problem-specific features, which significantly boost classifier accuracy and achieve above state-of-the-art results on recent benchmark datasets. This calls into question the value of using complex sophisticated models for stance classification without first doing informed feature extraction.

Leon Derczynski, Kalina Bontcheva, Maria Liakata et al.

Media is full of false claims. Even Oxford Dictionaries named "post-truth" as the word of 2016. This makes it more important than ever to build systems that can identify the veracity of a story, and the kind of discourse there is around it. RumourEval is a SemEval shared task that aims to identify and handle rumours and reactions to them, in text. We present an annotation scheme, a large dataset covering multiple topics - each having their own families of claims and replies - and use these to pose two concrete challenges as well as the results achieved by participants on these challenges.

Isabelle Augenstein, Leon Derczynski, Kalina Bontcheva

Named Entity Recognition (NER) is a key NLP task, which is all the more challenging on Web and user-generated content with their diverse and continuously changing language. This paper aims to quantify how this diversity impacts state-of-the-art NER methods, by measuring named entity (NE) and context variability, feature sparsity, and their effects on precision and recall. In particular, our findings indicate that NER approaches struggle to generalise in diverse genres with limited training data. Unseen NEs, in particular, play an important role, which have a higher incidence in diverse genres such as social media than in more regular genres such as newswire. Coupled with a higher incidence of unseen features more generally and the lack of large training corpora, this leads to significantly lower F1 scores for diverse genres as compared to more regular ones. We also find that leading systems rely heavily on surface forms found in training data, having problems generalising beyond these, and offer explanations for this observation.

Leon Derczynski

A plethora of vector-space representations for words is currently available, which is growing. These consist of fixed-length vectors containing real values, which represent a word. The result is a representation upon which the power of many conventional information processing and data mining techniques can be brought to bear, as long as the representations are designed with some forethought and fit certain constraints. This paper details desiderata for the design of vector space representations of words.

Leon Derczynski, Isabelle Augenstein, Kalina Bontcheva

This paper describes a pilot NER system for Twitter, comprising the USFD system entry to the W-NUT 2015 NER shared task. The goal is to correctly label entities in a tweet dataset, using an inventory of ten types. We employ structured learning, drawing on gazetteers taken from Linked Data, and on unsupervised clustering features, and attempting to compensate for stylistic and topic drift - a key challenge in social media text. Our result is competitive; we provide an analysis of the components of our methodology, and an examination of the target dataset in the context of this task.

Leon Derczynski, Diana Maynard, Giuseppe Rizzo et al.

Applying natural language processing for mining and intelligent information access to tweets (a form of microblog) is a challenging, emerging research area. Unlike carefully authored news text and other longer content, tweets pose a number of new challenges, due to their short, noisy, context-dependent, and dynamic nature. Information extraction from tweets is typically performed in a pipeline, comprising consecutive stages of language identification, tokenisation, part-of-speech tagging, named entity recognition and entity disambiguation (e.g. with respect to DBpedia). In this work, we describe a new Twitter entity disambiguation dataset, and conduct an empirical analysis of named entity recognition and disambiguation, investigating how robust a number of state-of-the-art systems are on such noisy texts, what the main sources of error are, and which problems should be further investigated to improve the state of the art.

Steven Bethard, Leon Derczynski, James Pustejovsky et al.

We describe the Clinical TempEval task which is currently in preparation for the SemEval-2015 evaluation exercise. This task involves identifying and describing events, times and the relations between them in clinical text. Six discrete subtasks are included, focusing on recognising mentions of times and events, describing those mentions for both entity types, identifying the relation between an event and the document creation time, and identifying narrative container relations.

Leon Derczynski, Hector Llorens, Naushad UzZaman

TimeML is an XML-based schema for annotating temporal information over discourse. The standard has been used to annotate a variety of resources and is followed by a number of tools, the creation of which constitute hundreds of thousands of man-hours of research work. However, the current state of resources is such that many are not valid, or do not produce valid output, or contain ambiguous or custom additions and removals. Difficulties arising from these variances were highlighted in the TempEval-3 exercise, which included its own extra stipulations over conventional TimeML as a response. To unify the state of current resources, and to make progress toward easy adoption of its current incarnation ISO-TimeML, this paper introduces TimeML-strict: a valid, unambiguous, and easy-to-process subset of TimeML. We also introduce three resources -- a schema for TimeML-strict; a validator tool for TimeML-strict, so that one may ensure documents are in the correct form; and a repair tool that corrects common invalidating errors and adds disambiguating markup in order to convert documents from the laxer TimeML standard to TimeML-strict.

Leon Derczynski, Richard Shaw, Ben Solway et al.

Question answering involves developing methods to extract useful information from large collections of documents. This is done with specialised search engines such as Answer Finder. The aim of Answer Finder is to provide an answer to a question rather than a page listing related documents that may contain the correct answer. So, a question such as "How tall is the Eiffel Tower" would simply return "325m" or "1,063ft". Our task was to build on the current version of Answer Finder by improving information retrieval, and also improving the pre-processing involved in question series analysis.

Naushad UzZaman, Hector Llorens, James Allen et al.

We describe the TempEval-3 task which is currently in preparation for the SemEval-2013 evaluation exercise. The aim of TempEval is to advance research on temporal information processing. TempEval-3 follows on from previous TempEval events, incorporating: a three-part task structure covering event, temporal expression and temporal relation extraction; a larger dataset; and single overall task quality scores.

Leon Derczynski, Jun Wang, Robert Gaizauskas et al.

Automated answering of natural language questions is an interesting and useful problem to solve. Question answering (QA) systems often perform information retrieval at an initial stage. Information retrieval (IR) performance, provided by engines such as Lucene, places a bound on overall system performance. For example, no answer bearing documents are retrieved at low ranks for almost 40% of questions. In this paper, answer texts from previous QA evaluations held as part of the Text REtrieval Conferences (TREC) are paired with queries and analysed in an attempt to identify performance-enhancing words. These words are then used to evaluate the performance of a query expansion method. Data driven extension words were found to help in over 70% of difficult questions. These words can be used to improve and evaluate query expansion methods. Simple blind relevance feedback (RF) was correctly predicted as unlikely to help overall performance, and an possible explanation is provided for its low value in IR for QA.

Leon Derczynski, Héctor Llorens, Estela Saquete

Automatic annotation of temporal expressions is a research challenge of great interest in the field of information extraction. Gold standard temporally-annotated resources are limited in size, which makes research using them difficult. Standards have also evolved over the past decade, so not all temporally annotated data is in the same format. We vastly increase available human-annotated temporal expression resources by converting older format resources to TimeML/TIMEX3. This task is difficult due to differing annotation methods. We present a robust conversion tool and a new, large temporal expression resource. Using this, we evaluate our conversion process by using it as training data for an existing TimeML annotation tool, achieving a 0.87 F1 measure -- better than any system in the TempEval-2 timex recognition exercise.

Amev Burman, Arun Jayapal, Sathish Kannan et al.

This paper describes the University of Sheffield's entry in the 2011 TAC KBP entity linking and slot filling tasks. We chose to participate in the monolingual entity linking task, the monolingual slot filling task and the temporal slot filling tasks. We set out to build a framework for experimentation with knowledge base population. This framework was created, and applied to multiple KBP tasks. We demonstrated that our proposed framework is effective and suitable for collaborative development efforts, as well as useful in a teaching environment. Finally we present results that, while very modest, provide improvements an order of magnitude greater than our 2010 attempt.