Neural organization of speech production: A lesion-based study of error patterns in connected speech

Brielle C Stark; Alexandra Basilakos; Gregory Hickok; Chris Rorden; Leonardo Bonilha; Julius Fridriksson

Much of what we know about the neural architecture of language production is grounded in studies of single word retrieval (Levelt, 1999; Foygel and Dell, 2000; Walker and Hickok, 2016) . To successfully retrieve a word, such as the name of an object, one must access at least three levels of information: conceptual (what it is), lexical (its associated word) and phonological (the sounds to select and organize). When we describe a situation, however, we do not simply name objects. Instead, we select two or three words per second from an active vocabulary spanning an estimated 40,000 words (Levelt, 1989) , and making this feat more complex, our word selection is vulnerable to "competition" from multiple sources. For example, errors in connected speech tend to be phonologically or semantically related word substitutions or omissions, the replacement of the intended target word with a word that we intended to use in a future production, or the perseveration on a word that was used recently (Lashley, 1951; Garrett, 1975) . Selecting a word in connected speech, therefore, requires a delicate balance of target word activation and suppression of alternatives, where alternatives come in the form of previously selected forms, subsequent forms in various stages of planning, and similar competitor forms, all coordinated across multiple hierarchical levels.