This paper reports on an experiment in formalizing and implementing “story-sifting” heuristics to automatically identify likely “interesting stories” from a large mass of events transpiring from the Chroniqueur procedural social simulation. Following preliminary experimentation with various concepts inspired by narratology, we proceeded to formally evaluate “unexpectedness” and “dramatic situations” as key heuristics to sample six stories from a simulation run and rendered each as two short stories, one written by a human author and the other by a LLM (in order to help distinguish the effect of narrative from story). These were scored by 24 participants for narrative appreciation. Inconclusive results about heuristics making any significant difference prompted self-reflection on what makes emergent narratives interesting (in a distinct way from traditional fiction), pointing both at “story effects” (some selected event chains constituting better story material) and “narrative effect” (better narrativization being able to cope with poor story material).

This paper introduces the first multi-lingual and multi-label classification model for implicit discourse relation recognition (IDRR). Our model, HArch, is evaluated on the recently released DiscoGeM 2.0 corpus and leverages hierarchical dependencies between discourse senses to predict probability distributions across all three sense levels in the PDTB 3.0 framework. We compare several pre-trained encoder backbones and find that RoBERTa-HArch achieves the best performance in English, while XLM-RoBERTa-HArch performs best in the multi-lingual setting. In addition, we compare our fine-tuned models against GPT-4o and Llama-4-Maverick using few-shot prompting across all language configurations. Our results show that our fine-tuned models consistently outperform these LLMs, highlighting the advantages of task-specific fine-tuning over prompting in IDRR. Finally, we report SOTA results on the DiscoGeM 1.0 corpus, further validating the effectiveness of our hierarchical approach.

We propose a novel multi-label classification approach to implicit discourse relation recognition (IDRR). Our approach features a multi-task model that jointly learns multi-label representations of implicit discourse relations across all three sense levels in the PDTB 3.0 framework. The model can also be adapted to the traditional single-label IDRR setting by selecting the sense with the highest probability in the multi-label representation. We conduct extensive experiments to identify optimal model configurations and loss functions in both settings. Our approach establishes the first benchmark for multi-label IDRR and achieves SOTA results on single-label IDRR using DiscoGeM. Finally, we evaluate our model on the PDTB 3.0 corpus in the single-label setting, presenting the first analysis of transfer learning between the DiscoGeM and PDTB 3.0 corpora for IDRR.

In this paper, we propose an approach to automate the process of place name detection in the medical domain to enable epidemiologists to better study and model the spread of viruses.We created a family of Toponym Identification Models based on BERT (TIMBERT), in order to learn in an end-to-end fashion the mapping from an input sentence to the associated sentence labeled with toponyms.When evaluated with the SemEval 2019 task 12 test set (Weissenbacher et al., 2019), our best TIMBERT model achieves an F1 score of 90.85%, a significant improvement compared to the state-of-the-art of 89.

In this paper, we report on our experiments towards the creation of a corpus for coherence evaluation. Most corpora for textual coherence evaluation are composed of randomly shuffled sentences that focus on sentence ordering, regardless of whether the sentences were originally related by a discourse relation. To the best of our knowledge, no publicly available corpus has been designed specifically for the evaluation of coherence of known discursive units. In this paper, we focus on coherence modeling at the intra-discursive level and describe our approach to build a corpus of incoherent pairs of sentences. We experimented with a variety of corruption strategies to create synthetic incoherent pairs of discourse arguments from coherent ones. Using discourse argument pairs from the Penn Discourse Tree Bank, we generate incoherent discourse argument pairs, by swapping either their discourse connective or a discourse argument. To evaluate how incoherent the generated corpora are, we use a convolutional neural network to try to distinguish the original pairs from the corrupted ones. Results of the classifier as well as a manual inspection of the corpora show that generating such corpora is still a challenge as the generated instances are clearly not “incoherent enough”, indicating that more effort should be spent on developing more robust ways of generating incoherent corpora.

In this paper, we propose a neural-based model to address the first task of the DEFT 2013 shared task, with the main challenge of a highly imbalanced dataset, using state-of-the-art embedding approaches and deep architectures. We report on our experiments on the use of linguistic features, extracted by Charton et. al. (2014), in different neural models utilizing pretrained embeddings. Our results show that all of the models that use linguistic features outperform their counterpart models that only use pretrained embeddings. The best performing model uses pretrained CamemBERT embeddings as input and CNN as the hidden layer, and uses additional linguistic features. Adding the linguistic features to this model improves its performance by 4.5% and 11.4% in terms of micro and macro F1 scores, respectively, leading to state-of-the-art results and an improved classification of the rare classes.

In this paper, we present an approach to exploit phrase tables generated by statistical machine translation in order to map French discourse connectives to discourse relations.Using this approach, we created ConcoLeDisCo, a lexicon of French discourse connectives and their PDTB relations.When evaluated against LEX-CONN, ConcoLeDisCo achieves a recall of 0.81 and an Average Precision of 0.68 for the CONCESSION and CONDITION relations.

This paper investigates the influence of discourse features on text complexity assessment.To do so, we created two data sets based on the Penn Discourse Treebank and the Simple English Wikipedia corpora and compared the influence of coherence, cohesion, surface, lexical and syntactic features to assess text complexity.Results show that with both data sets coherence features are more correlated to text complexity than the other types of features.In addition, feature selection revealed that with both data sets the top most discriminating feature is a coherence feature.

This paper presents a supervised learning approach to support the screening of HIV literature. The manual screening of biomedical literature is an important task in the process of systematic reviews. Researchers and curators have the very demanding, time-consuming and error-prone task of manually identifying documents that must be included in a systematic review concerning a specific problem. We implemented a supervised learning approach to support screening tasks, by automatically flagging potentially selected documents in a list retrieved by a literature database search. To overcome the main issues associated with the automatic literature screening task, we evaluated the use of data sampling, feature combinations, and feature selection methods, generating a total of 105 classification models. The models yielding best results were composed by the Logistic Model Trees classifier, a fairly balanced training set, and feature combination of Bag-Of-Words and MeSH terms. According to our results, the system correctly labels the great majority of relevant documents, and it could be used to support HIV systematic reviews to allow researchers to assess a greater number of documents in less time.

In this paper we propose a method for semantic inter-media photographic retrieval, exploiting the advantages of both textual and content-based frameworks: the relatively high initial precision and diversity of results from visual and text retrieval, and the robust overall precision and recall of text retrieval. The method employs simple block-based visual retrieval which, at early precision, outperforms MPEG-7 features ScalableColor, ColorLayout and EdgeHistogram, allowing for reliable auto-relevance feedback. The query expansion applied respects the semantic constraints of the original query and enhances precision and confidence through redundancy. The proposed method was tested on a benchmark data set of 20,000 diverse tourist photographs and obtained promising results.

This paper discusses a fundamental problem in natural language generation: how to organize the content of a text in a coherent and natural way. In this research, we set out to determine the semantic content and the rhetorical structure of texts and to develop heuristics to perform this process automatically within a text generation framework. The study was performed on a specific language and textual genre: French instructional texts. From a corpus analysis of these texts, we determined nine senses typically communicated in instructional texts and seven rhetorical relations used to present these senses. From this analysis, we then developed a set of presentation heuristics that determine how the senses to be communicated should be organized rhetorically in order to create a coherent and natural text. The heuristics are based on five types of constraints: conceptual, semantic, rhetorical, pragmatic, and intentional constraints. To verify the heuristics, we developed the spin natural language generation system, which performs all steps of text generation but focuses on the determination of the content and the rhetorical structure of the text.