Rethinking the Mental Lexicon

In class and our reading, the term lexicon has been briefly introduced. The lexicon is the “mental dictionary”—a data structure that stores knowledge about words. I’m fascinated by the inner workings of the brain, so I found myself curious about what is actually stored in this structure. This led me to do a little general research about the lexicon, and I’ve found an interesting new way to think about the mental lexicon. This is according to the ideas of Jeffrey Elman, a well-known psycholinguist at UC San Diego. Elman takes particular interest in neural networks and has used this idea to rethink not only what information the mental lexicon stores, but what kind of structure the mental lexicon has.

           

The Conflict

            Recent studies have shown that event knowledge and expectations play a significant role in the early stages of sentence processing. Many dynamic factors, which have been considered to lie outside of the lexicon, such as a verb’s grammatical aspect (how an action relates to the flow of time), the agent and instrument involved in an activity, and the overall discourse context, have been shown to significantly affect expectations.

            It has been shown that all of this information, including from dynamic factors, is available in the same time frame, while time frame has been used to determine operationally what is stored in the lexicon and what is stored elsewhere. Therefore, Elman argues that the dictionary-style mental lexicon is insufficient because dynamic factors bring in too much context-specific information to possibly be contained in the mental lexicon while still maintaining a distinction between the mental lexicon and other linguistic modules.

Rethinking the Mental Lexicon

Elman suggests an alternative: lexical knowledge without a lexicon. Elman describes his new way of thinking about the lexicon as this: “Alternatively (but equivalently) one can view words not as elements in a data structure that must be retrieved from memory, but rather as stimuli that alter mental states (which arise from processing prior words) in lawful ways. In this view, words are not mental objects that reside in a mental lexicon. They are operators on mental states.”

By saying words are “stimuli that alter mental states (which arise from processing prior words),” Elman is trying to take into account the dynamic factors that studies have shown that the mental lexicon has quick access to. 

The Simple Recurrent Network

            A simple system that could model the context-dependent nature of lexical knowledge is a connectionist model called a simple recurrent network. In this model, context information for what has been previously said is added for each step in time to the overall comprehension. This constant connection to context helps explain how studies have shown that the mental lexicon has such quick access to dynamic factors, like context.

            But how does this network understand context in the first place? The network makes context-dependent predictions that approximate the probabilities of succeeding elements, that is, one word leads to the next by probability determined from context instead of some kind of information being stored with a word (like its definition), as in the traditional lexicon view.

To do this, the network maps in multidimensional space connections between words. This movement from word to word can be viewed as a trajectory of words. Multiple meanings of words are not stored in the same place—this explains nuanced, contextual understanding. For example, when the agent butcher is said before saw, the trajectory leads to a specific understanding of the word saw based on world knowledge (one that will be different from the understanding of saw after the word lumberjack, for instance).

Source

Elman, J. (2011). Lexical knowledge without a lexicon?. The Mental Lexicon, 1-33.