Commonsense is a quintessential human capacity that has been a core challenge to Artificial Intelligence since its inception. Impressive results in Natural Language Processing tasks, including in commonsense reasoning, have consistently been achieved with Transformer neural language models, even matching or surpassing human performance in some benchmarks. Recently, some of these advances have been called into question: so called data artifacts in the training data have been made evident as spurious correlations and shallow shortcuts that in some cases are leveraging these outstanding results. In this paper we seek to further pursue this analysis into the realm of commonsense related language processing tasks. We undertake a study on different prominent benchmarks that involve commonsense reasoning, along a number of key stress experiments, thus seeking to gain insight on whether the models are learning transferable generalizations intrinsic to the problem at stake or just taking advantage of incidental shortcuts in the data items. The results obtained indicate that most datasets experimented with are problematic, with models resorting to non-robust features and appearing not to be learning and generalizing towards the overall tasks intended to be conveyed or exemplified by the datasets.

Lexical semantics theories differ in advocating that the meaning of words is represented as an inference graph, a feature mapping or a cooccurrence vector, thus raising the question: is it the case that one of these approaches is superior to the others in representing lexical semantics appropriately? Or in its non antagonistic counterpart: could there be a unified account of lexical semantics where these approaches seamlessly emerge as (partial) renderings of (different) aspects of a core semantic knowledge base? In this paper, we contribute to these research questions with a number of experiments that systematically probe different lexical semantics theories for their levels of cognitive plausibility and of technological usefulness. The empirical findings obtained from these experiments advance our insight on lexical semantics as the feature-based approach emerges as superior to the other ones, and arguably also move us closer to finding answers to the research questions above.

Reproduction of scientific findings is essential for scientific development across all scientific disciplines and reproducing results of previous works is a basic requirement for validating the hypothesis and conclusions put forward by them. This paper reports on the scientific reproduction of several systems addressing the Argument Reasoning Comprehension Task of SemEval2018. Given a recent publication that pointed out spurious statistical cues in the data set used in the shared task, and that produced a revised version of it, we also evaluated the reproduced systems with this new data set. The exercise reported here shows that, in general, the reproduction of these systems is successful with scores in line with those reported in SemEval2018. However, the performance scores are worst than those, and even below the random baseline, when the reproduced systems are run over the revised data set expunged from data artifacts. This demonstrates that this task is actually a much harder challenge than what could have been perceived from the inflated, close to human-level performance scores obtained with the data set used in SemEval2018. This calls for a revival of this task as there is much room for improvement until systems may come close to the upper bound provided by human performance.

An effective conversion method was proposed in the literature to obtain a lexical semantic space from a lexical semantic graph, thus permitting to obtain WordNet embeddings from WordNets. In this paper, we propose the exploitation of this conversion methodology as the basis for the comparative assessment of WordNets: given two WordNets, their relative quality in terms of capturing the lexical semantics of a given language, can be assessed by (i) converting each WordNet into the corresponding semantic space (i.e. into WordNet embeddings), (ii) evaluating the resulting WordNet embeddings under the typical semantic similarity prediction task used to evaluate word embeddings in general; and (iii) comparing the performance in that task of the two word embeddings, extracted from the two WordNets. A better performance in that evaluation task results from the word embeddings that are better at capturing the semantic similarity of words, which, in turn, result from the WordNet that is of higher quality at capturing the semantics of words.

Vectorial representations of meaning can be supported by empirical data from diverse sources and obtained with diverse embedding approaches. This paper aims at screening this experimental space and reports on an assessment of word embeddings supported (i) by data in raw texts vs. in lexical graphs, (ii) by lexical information encoded in association- vs. inference-based graphs, and obtained (iii) by edge reconstruction- vs. matrix factorisation vs. random walk-based graph embedding methods. The results observed with these experiments indicate that the best solutions with graph-based word embeddings are very competitive, consistently outperforming mainstream text-based ones.

The task of taking a semantic representation of a noun and predicting the brain activity triggered by it in terms of fMRI spatial patterns was pioneered by Mitchell et al. 2008. That seminal work used word co-occurrence features to represent the meaning of the nouns. Even though the task does not impose any specific type of semantic representation, the vast majority of subsequent approaches resort to feature-based models or to semantic spaces (aka word embeddings). We address this task, with competitive results, by using instead a semantic network to encode lexical semantics, thus providing further evidence for the cognitive plausibility of this approach to model lexical meaning.

Semantic networks and semantic spaces have been two prominent approaches to represent lexical semantics. While a unified account of the lexical meaning relies on one being able to convert between these representations, in both directions, the conversion direction from semantic networks into semantic spaces started to attract more attention recently. In this paper we present a methodology for this conversion and assess it with a case study. When it is applied over WordNet, the performance of the resulting embeddings in a mainstream semantic similarity task is very good, substantially superior to the performance of word embeddings based on very large collections of texts like word2vec.

This paper presents the results of systematic experimentation on the impact in duplicate question detection of different types of questions across both a number of established approaches and a novel, superior one used to address this language processing task. This study permits to gain a novel insight on the different levels of robustness of the diverse detection methods with respect to different conditions of their application, including the ones that approximate real usage scenarios.