A model of early word acquisition based on realistic-scale audiovisual naming events

As they grow, infants gradually acquire understanding of their native language without direct supervision. By the age of 6 months, infants' perception has already attuned to native language phonetic contrasts [1] and they show first signs of word comprehension [2-4] and familiar word identification [5-7]. By 12 months, they already recognize dozens of words [8]. During this learning process, the infants must learn to parse the speech stream into words and to associate the words with their referential meanings in the external world. From a cognitive perspective, the discovery of words and word-meaning mappings is a task far from trivial: acoustic speech is a continuous and complex signal without transparent linguistic structure (see, e.g., [9, 10]), and there is substantial ambiguity in how individual words embedded in larger utterances are related to specific objects and events in the visual scene, also known as referential ambiguity [11]. Figure 1 illustrates referential ambiguity in everyday life situations between speech and its visual references through some examples.