RESUMO
The constructivist acquisition of language by children has been elaborately documented by researchers in psycholinguistics and cognitive science. However, despite the centrality of human-like communication in the field of artificial intelligence, no faithful computational operationalizations of the mechanisms through which children learn language exist to date. In this article, we fill part of this void by introducing a mechanistic model of the constructivist acquisition of language through syntactico-semantic pattern finding. Concretely, we present a methodology for learning grammars based on similarities and differences in the form and meaning of linguistic observations alone. The resulting grammars consist of form-meaning mappings of variable extent and degree of abstraction, called constructions, which facilitate both language comprehension and production. Applying our methodology to the CLEVR benchmark dataset, we provide a proof of concept that demonstrates the online, incremental, data-efficient, transparent and effective learning of item-based construction grammars from utterance-meaning pairs.
RESUMO
Since its inception in the mid-eighties, the field of construction grammar has been steadily growing and constructionist approaches to language have by now become a mainstream paradigm for linguistic research. While the construction grammar community has traditionally focused on theoretical, experimental and corpus-based research, the importance of computational methodologies is now rapidly increasing. This movement has led to the establishment of a number of exploratory computational construction grammar formalisms, which facilitate the implementation of construction grammars, as well as their use for language processing purposes. Yet, implementing large grammars using these formalisms still remains a challenging task, partly due to a lack of powerful and user-friendly tools for computational construction grammar engineering. In order to overcome this obstacle, this paper introduces the FCG Editor, a dedicated and innovative integrated development environment for the Fluid Construction Grammar formalism. Offering a straightforward installation and a user-friendly, interactive interface, the FCG Editor is an accessible, yet powerful tool for construction grammarians who wish to operationalise their construction grammar insights and analyses in order to computationally verify them, corroborate them with corpus data, or integrate them in language technology applications.
Assuntos
Idioma , LinguísticaRESUMO
With more and more voices and opinions entering the public domain, a key challenge facing journalists and editors is maximizing the context of the information that is presented on news websites. In this paper, we argue that systems for exposing readers to the many aspects of societal debates should be grounded in methods and tools that can provide a fine-grained understanding of these debates. The present article thereby explores the conceptual transition from opinion observation to opinion facilitation by introducing and discussing the Penelope opinion facilitator: a proof-of-concept reading instrument for online news media that operationalizes emerging methods for the computational analysis of cultural conflict developed in the context of the H2020 ODYCCEUS project. It will be demonstrated how these methods can be combined into an instrument that complements the reading experience of the news website The Guardian by automatically interlinking news articles on the level of semantic frames. In linguistic theory, semantic frames are defined as coherent structures of related concepts. We thereby zoom in on instances of the "causation" frame, such as "climate change causes global warming," and illustrate how a reading instrument that links articles based on such frames might reconfigure our readings of climate news coverage, with specific attention to the case of global warming controversies. Finally, we relate our findings to the context of the development of computational social science, and discuss pathways for the evaluation of the instrument, as well as for the future upscaling of qualitative analyses and close readings.
RESUMO
Autonomous agents perceive the world through streams of continuous sensori-motor data. Yet, in order to reason and communicate about their environment, agents need to be able to distill meaningful concepts from their raw observations. Most current approaches that bridge between the continuous and symbolic domain are using deep learning techniques. While these approaches often achieve high levels of accuracy, they rely on large amounts of training data, and the resulting models lack transparency, generality, and adaptivity. In this paper, we introduce a novel methodology for grounded concept learning. In a tutor-learner scenario, the method allows an agent to construct a conceptual system in which meaningful concepts are formed by discriminative combinations of prototypical values on human-interpretable feature channels. We evaluate our approach on the CLEVR dataset, using features that are either simulated or extracted using computer vision techniques. Through a range of experiments, we show that our method allows for incremental learning, needs few data points, and that the resulting concepts are general enough to be applied to previously unseen objects and can be combined compositionally. These properties make the approach well-suited to be used in robotic agents as the module that maps from continuous sensory input to grounded, symbolic concepts that can then be used for higher-level reasoning tasks.