Recently, several specialized instruction-tuned Large Language Models (LLMs) for Named Entity Recognition (NER) have emerged. Compared with traditional NER approaches, these models have demonstrated strong generalization capabilities. Existing LLMs primarily focus on addressing zero-shot NER on Out-of-Domain inputs, while fine-tuning on an extensive number of entity classes that often highly or completely overlap with test sets. In this work instead, we propose SLIMER, an approach designed to tackle never-seen-before entity tags by instructing the model on fewer examples, and by leveraging a prompt enriched with definition and guidelines. Experiments demonstrate that definition and guidelines yield better performance, faster and more robust learning, particularly when labelling unseen named entities. Furthermore, SLIMER performs comparably to state-of-the-art approaches in out-of-domain zero-shot NER, while being trained in a more fair, though certainly more challenging, setting.

Crossword puzzles are popular linguistic games often used as tools to engage students in learning. Educational crosswords are characterized by less cryptic and more factual clues that distinguish them from traditional crossword puzzles. Despite there exist several publicly available clue-answer pair databases for traditional crosswords, educational clue-answer pairs datasets are missing. In this article, we propose a methodology to build educational clue generation datasets that can be used to instruct Large Language Models (LLMs). By gathering from Wikipedia pages informative content associated with relevant keywords, we use Large Language Models to automatically generate pedagogical clues related to the given input keyword and its context. With such an approach, we created clue-instruct, a dataset containing 44,075 unique examples with text-keyword pairs associated with three distinct crossword clues. We used clue-instruct to instruct different LLMs to generate educational clues from a given input content and keyword. Both human and automatic evaluations confirmed the quality of the generated clues, thus validating the effectiveness of our approach.

To reduce the inference cost of large language models, model compression is increasingly used to create smaller scalable models.However, little is known about their robustness to minority subgroups defined by the labels and attributes of a dataset.In this paper, we investigate the effects of 18 different compression methods and settings on the subgroup robustness of BERT language models.We show that worst-group performance does not depend on model size alone, but also on the compression method used.Additionally, we find that model compression does not always worsen the performance on minority subgroups.Altogether, our analysis serves to further research into the subgroup robustness of model compression.

Transformers have been established as one of the most effective neural approach in performing various Natural Language Processing tasks. However, following common trend in modern deep architectures, their scale has quickly grown to an extent that reduces the concrete possibility for several enterprises to train such models from scratch. Indeed, despite their high-level performances, Transformers have the general drawback of requiring a huge amount of training data, computational resources and energy consumption to be successfully optimized. For this reason, more recent architectures like Bidirectional Encoder Representations from Transformers rely on unlabeled data to pre-train the model, which is later fine-tuned for a specific downstream task using a relatively smaller amount of training data. In a similar fashion, this paper considers a plug-and-play framework that can be used to inject multiple syntactic features, like Part-of-Speech Tagging or Dependency Parsing, into any kind of pre-trained Transformer. This novel approach allows to perform sequence-to-sequence labeling tasks by exploiting: (i) the (more abundant) available training data that is also used to learn the syntactic features, (ii) the language data that is used to pre-train the transformer model. The experimental results show that our approach improves over the baseline performances of the underlying model in different datasets, thus proving the effectiveness of employing syntactic language information for semantic regularization. In addition, we show that our architecture has a huge efficiency advantage over pure large language models. Indeed, by using a model with limited size, but whose input data are enriched with syntactic information, we show that it is possible to obtain a significant reduction of CO2 emissions without decreasing the prediction performances.

In the last decades, deep learning approaches achieved impressive results in many research fields, such as Computer Vision and Natural Language Processing (NLP). NLP in particular has greatly benefit from unsupervised methods that allow to learn distributed representation of language. On the race for better performances Language Models have reached hundred of billions parameters nowadays. Despite the remarkable results, deep models are still far from being fully exploited in real world applications. Indeed, these approaches are black-boxes, i.e. they are not interpretable by design nor explainable, which is often crucial to make decisions in business. Several task-agnostic methods have been proposed in literature to explain models' decisions. Most techniques rely on the “local” assumption, i.e. explanations are made example-wise. In this paper instead, we present a post-hoc method to produce highly interpretable global rules to explain NLP classifiers. Rules are extracted with a data mining approach on a semantically enriched input representation, instead of using words/wordpieces solely. Semantic information yields more abstract and general rules that are both more explanatory and less complex, while being also better at reflecting the model behaviour. In the experiments we focus on Named Entity Recognition, an NLP task where explainability is under-investigated. We explain the predictions of BERT NER classifiers trained on two popular benchmarks, CoNLL03 and Ontonotes, and compare our model against LIME [1] and Decision Trees.

Writers, poets, singers usually do not create their compositions in just one breath. Text is revisited, adjusted, modified, rephrased, even multiple times, in order to better convey meanings, emotions and feelings that the author wants to express. Amongst the noble written arts, Poetry is probably the one that needs to be elaborated the most, since the composition has to formally respect predefined meter and rhyming schemes. In this paper, we propose a framework to generate poems that are repeatedly revisited and corrected, as humans do, in order to improve their overall quality. We frame the problem of revising poems in the context of Reinforcement Learning and, in particular, using Proximal Policy Optimization. Our model generates poems from scratch and it learns to progressively adjust the generated text in order to match a target criterion. We evaluate this approach in the case of matching a rhyming scheme, without having any information on which words are responsible of creating rhymes and on how to coherently alter the poem words. The proposed framework is general and, with an appropriate reward shaping, it can be applied to other text generation problems.

We consider a scenario where an artificial agent is reading a stream of text composed of a set of narrations, and it is informed about the identity of some of the individuals that are mentioned in the text portion that is currently being read. The agent is expected to learn to follow the narrations, thus disambiguating mentions and discovering new individuals. We focus on the case in which individuals are entities and relations and propose an end-to-end trainable memory network that learns to discover and disambiguate them in an online manner, performing one-shot learning and dealing with a small number of sparse supervisions. Our system builds a not-given-in-advance knowledge base, and it improves its skills while reading the unsupervised text. The model deals with abrupt changes in the narration, considering their effects when resolving coreferences. We showcase the strong disambiguation and discovery skills of our model on a corpus of Wikipedia documents and on a newly introduced data set that we make publicly available.