Hypotheses about the nature and content of the mental lexicon have played a prominent part in many of the most important debates in phonological theory in the past half century. Diametrically opposed arguments have been advanced concerning how much phonological and phonetic information may be contained in lexical representations; whether the lexicon is computationally inert or rather provides the domain for significant synchronic generalizations; and to what extent lexical entries are autonomous from, or overlap with, other types of representations in long-term memory.

Early generative phonology espoused a set of tightly interlocking assumptions concerning the mental lexicon. In line with the structuralist tradition, the theory denied the existence of word-specific phonetic detail. Going further, the evaluation measure assigned the same cost to symbol tokens in underlying representations and in the statement of phonological rules. This resulted in a maximally economical lexicon, with all redundant information introduced in the rule component.

In the 1970s and 1980s, the assumption of underspecified lexical representations was put to use to explain a range of phenomena, such as lexically diffusing change or the fact that redundant features are often phonologically inert. This use of underspecification has lately fallen largely out of favour, but there is a line of work that adduces psycholinguistic evidence from word recognition phenomena in favour of underspecified lexical entries.

In compliance with the principle of Richness of the Base, classic OT has abandoned lexical underspecification. However, Richness of the Base upholds, and indeed strengthens, other fundamental assumptions of early generative models. All systematic phonological patterns are captured by means of constraint ranking, with the bias for markedness over faithfulness replacing SPE's evaluation measure. If language-particular generalizations hold over lexical entries at all, they do so inertly and indirectly by virtue of Lexicon Optimization.

Examplar theory departs from the structuralist tradition in a much more dramatic fashion. In its most radical versions, this theory depicts phonetic detail as lexically specific, reduces phonological generalizations to patterns inherent in memory traces, and abolishes the modular separation of the mental lexicon from episodic long-term memory.

The special session will explore the state of the art on these fundamental issues.