• analogy
• compositionality
• Construction Grammar
• Distributional Models of Meaning
• language comprehension
• Usage-based theories

In this dissertation, we explore two lines of research framed within the usage-based constructionist paradigm. On one side, we investigate how to ground the semantic content of constructions in language use; we propose integrating vector representations used in Distributional Semantic Models into linguistic descriptions of Construction Grammar.

Besides, we address a still open question: What cognitive and linguistic principles govern language comprehension? Considerable evidence suggests that interpretation alternates compositional (incremental) and noncompositional (global) strategies. Although it is recognized that idioms are fast to process, we claim that even literal expressions, if frequent enough, are processed in the same way.
Using the Self-Paced Reading paradigm, we tested reading times of idiomatic and literal high-frequent and low-frequent verb-noun phrases, observing that facilitation effects also occur when processing frequent and yet compositional expressions.

Concurrently, we claim that systematic processes of language productivity are mainly explainable by analogical inferences rather than sequential compositional operations: novel expressions are produced and understood `on the fly' by analogy with familiar ones.
As the principle of compositionality has been used to generate distributional representations of phrases, we propose a neural network simulating the construction of phrasal embedding as an analogical process. Our ANNE, inspired by word2vec and computer vision techniques, was evaluated on its ability to generalize from existing vectors.

Overall, we hope this work could clarify the complex literature on language comprehension and pave the way for new experimental and computational studies