Marco Marelli
University of Milano-Bicocca
Main areas of research
- Psycholinguistics
- Cognitive science of meaning
- Computational semantics
- Language processing and reading
- Conceptual representation
Bio
Marco Marelli is a full professor of General Psychology at the University
of Milano-Bicocca, Milano, Italy. His works focus on the psychology of
language and memory, and in particular on the varying involvement of
semantics in a number of different processes, and the interface between
language, sensorimotor experience and the conceptual system. In his
research he combines methods from experimental psychology, cognitive
neuroscience and computational linguistics. He is currently principal
investigator of the ERC project “BraveNewWord”, dedicated to the semantics
of unfamiliar words as emerging from compositionality (at the level of both
sentences and morphologically complex words) and systematicity (in a
statistical learning perspective).
Keynote Lecture
Systematicity and morphology in the processing of unfamiliar words
We learn new words almost on a daily basis: as adults, a new element is introduced in our vocabulary every other day. With new words, we also learn about new objects and ideas – this steady lexical learning seems to constitute a most important source of expansion for the semantic system. However, when we experience, as adults, a new word, its referent is not typically immediately available in the same context. How then can we comprehend novel words, simply on the basis of an unfamiliar sequence of sounds or graphical elements?
Reliable distributional associations between the form of a word and its meaning seems to be the main driving force (along with linguistic context) behind such remarkable ability. Such systematicity is a prominent property of natural languages, and speakers are shown to largely exploit it via statistical learning. While lexical morphology is the most obvious and well-described example of systematicity, it arguably represents the tip of the proverbial iceberg within a more nuanced, continuous, and pervasive phenomenon.
In this talk I will address the role of systematic and morphological patterns in the comprehension of novel words in a computational psycholinguistic perspective, moving from the tenets of distributional semantics and applying a series of models that induce meaning representations by leveraging on sublexical elements. Through such an approach, one can obtain predictions concerning the meaning of unfamiliar words, that can in turn be tested in psycholinguistic experiments. I will show how model predictions align with behavioral data from a number of studies, providing meaningful estimates for morphologically complex and simple novel words. I will further address which role for morphology remains, once lexicalist assumptions are abandoned and models are allowed to rely on the semantic contribution of any possible sublexical unit.
In conclusion, the talk will show how systematicity plays a central role in how humans access meaning, and how sublexical elements guide the processing of unfamiliar words. Distributional models that are enriched by such sublexical information represent a useful tool for psycholinguistic investigations in this domain.