Language in a (Search) Box: Grounding Language Learning in Real-World Human-Machine Interaction

We investigate grounded language learning through real-world data, by\nmodelling a teacher-learner dynamics through the natural interactions occurring\nbetween users and search engines; in particular, we explore the emergence of\nsemantic generalization from unsupervised dense representations outside of\nsynthetic environments. A grounding domain, a denotation function and a\ncomposition function are learned from user data only. We show how the resulting\nsemantics for noun phrases exhibits compositional properties while being fully\nlearnable without any explicit labelling. We benchmark our grounded semantics\non compositionality and zero-shot inference tasks, and we show that it provides\nbetter results and better generalizations than SOTA non-grounded models, such\nas word2vec and BERT.\n