Linguistic structures are a key information source in semantic knowledge acquisition. Indeed, both adults and children (Bloom, 2000) can learn a new property of a known or unknown concept from a single linguistic exposure. We formalize this as the ability to 1) decide whether the structure encountered expresses a property relevant to the concept (“oomparumpuses have tails” is likely relevant, while “that oomparumpus is hungry” may not be); 2) assign a type to properties deemed relevant (tails are parts, while being a bird describes class membership). We investigate, computationally, to what extent a single exposure to a concept-property pair in linguistic context is sufficient for effective learning, and what prior knowledge makes this possible.
Modelling Semantic Property Acquisition from Single Linguistic Exposure
LENCI, ALESSANDRO;
2008-01-01
Abstract
Linguistic structures are a key information source in semantic knowledge acquisition. Indeed, both adults and children (Bloom, 2000) can learn a new property of a known or unknown concept from a single linguistic exposure. We formalize this as the ability to 1) decide whether the structure encountered expresses a property relevant to the concept (“oomparumpuses have tails” is likely relevant, while “that oomparumpus is hungry” may not be); 2) assign a type to properties deemed relevant (tails are parts, while being a bird describes class membership). We investigate, computationally, to what extent a single exposure to a concept-property pair in linguistic context is sufficient for effective learning, and what prior knowledge makes this possible.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
