His major works are Metaphors We Live By (with Mark Johnson, 1980) and Women, Fire, and Dangerous Things: What Categories Reveal About The Mind (1992)
From The Metaphorical Structure of the Human Conceptual System (contained in Perspectives on Cognitive Science, edited by Donald O. Norman):
In summary, abstract concepts are not defined by necessary and sufficient conditions. Instead they are defined by clusters of metaphors. Each metaphor gives a partial definition. These partial definitions overlap in certain ways, but in general they are inconsistent, and typically have inconsistent ontologies. Elsewhere we have given an elaborate theoretical and empirical account of metaphorical definition (Lakoff & Johnson, 1980), but here we would only like to stress that the usual concept of definition in terms of necessary and sufficient conditions will not do.
And later, concluding the same work:
The way ordinary people deal implicitly with the limitations of any one metaphor is by having many metaphors for comprehending different aspects of the same concept. As we saw, people in our culture have many different metaphors for IDEAS and the MIND, some of which are elaborate in one or another branch of Psychology and some of which are not. These clusters of metaphors serve the purpose of understanding better than any single metaphor could-even though they are partial and very often inconsistent with each other. Scientists, however, have tended to insist on complete and consistent theories. While consistency is generally desirable, there are times when it does not best serve the purpose of understanding. In particular, the insistence on maintaining a consistent extension of one metaphor may blind us to aspects of reality that are ignored or hidden by that metaphor. We would lke to suggest that there are times when scientific understanding may best be served by permitting alternative metaphors even at the expense of completeness and consistency. If Cognitive Science is to be concerned wtih human understanding in its full richness, and not merely with those phenomena that fit the MIND IS A MACHINE metaphor, then it may have to sacrifice metaphorical consistency in the service of fuller understanding. The moral: Cognitive Science needs to be aware of its metaphors, to be concerned with what they hide, and to be open to alternative metaphors-even if they are inconsistent with the current favorites.
Seemingly paradoxical? In this context, consider Don Norman's paraphrase of Minsky: "After all, would you rather a scientific theory be correct or would you rather it be intelligible?"
(More will follow when I get more time to scan in the diagram referred to in the next paragraph.)
Cognitive Models and Cognitive Semantics:
Lakoff has approached the subject of grammar and semantics in a way that appears to be more in accord with the biological and psychological facts than are generative grammars. He starts from actual data on categorization and proposes that meaning results from the intrinsic workings of the body and the brain. He suggests that individual humans construct cognitive models that reflect concepts concerned with the interactions between the body-brain and the environment. It is this conceptual embodiment, he claims, that leads to the formulation of basic-level categories of the kind described by Rosch.
Returning to Biology
Cognitive models are created by human beings, and in this sense they are idealized-that is, they are abstractions. But they depend on the formation of images as a result of sensory experience and they also depend on kinesthetic experience-the relation of the body to space. Lakoff suggests that the exercise of these functions leads to various image and kinesthetic schemas. Schemas have properties that are reflected later in the use of metaphor and metonyms. Recall that a metaphor is the referral or mapping of one thing to another in a different domain, while a metonym is the use of some part or aspect of a thing to stand for the thing itself. Lakoff's example of a metaphor is "Anger is a dangerous animal." His example of a metonym is "The ham sandwich left without paying."
The important thing to grasp is that idealized cognitive models involve conceptual embodiment and that conceptual embodiment occurs through bodily activities prior to language. Conceptual embodiment is used in categorization and it allows for the heterogeneity and complexity of real human categorization. Categories of mind correspond accordingly to elements in cognitive models. Some of these models have different degrees of membership. Others include classical categories and are formed accoriding to singly necessary and jointly sufficient conditions (note that there is no contradiction here, provided that not all the models are classical!). Some models are metonymic. But the most complex cognitive models correspond to what Lakoff calls radial categories. These consist of many models linked around a center; they are saidby Lakoff to be 'motivated" by it.