On lexical minimalism
Many theories of lexical representation assume that the purpose of lexical representations is to account for the “meaning” of words, and suppose that, to do this, it is necessary to have a rather rich set of lexical specifications for each lexeme. Models as different as those of Jackendoff (e. g. Jackendoff, 1990), Wierzbicka (e. g. Wierzbicka, 199?), Pustejovsky (e. g. Pustejovsky, 1995), Katz, Katz and Fodor, and many others agree about this. There is, in general, little or no attention to the possibility that so much of the knowledge we use to understand words in context might be represented elsewhere that there is no need for more than a quite minimal set of specifications for a lexeme, most of the information needed for its use being specified at other levels.
An alternative model has seemed to me to be worth developing, and I have explored some of its aspects. On this model, which might be called the “minimal lexicon model”, the only specifications a language learner associates with a lexeme are specifications needed to distinguish that lexeme from others. This principle goes back, of course, to the structuralists, notably Saussure, and was adopted in Katz and Fodor’s 1963 model. What makes all such proposals different from one another is rather obvious. What one thinks needs to be specified at the lexical entry depends on what one thinks can be specified elsewhere. Some theoreticians (e. g. Pustejovsky, Wierzbicka) avoid speculation about the conceptual level, and it is not surprising that their lexical entries tend to be much fatter than those of theoreticians (like Jackendoff) who do take the conceptual level into account. Other theoreticians (like Langacker) reject the distinction between the lexical and conceptual levels, and even between knowledge of and knowledge about (encyclopedic knowledge) a concept. But even among theoreticians who distinguish between lexical and conceptual levels, what they assign to the lexical entry depends crucially on how they view the conceptual level. My view is that most of the specifications we use in understanding lexemes do not come from the lexicon: they have multiple sources, including idiosyncratic (concept-specific) specifications at different levels, and many more general sources. I take as a methodological principle that all information should be specified only once in the RD, but this leaves it open where it is specified.
In cases like “horse”, discussed in Carter, 1998, since there exists a sizable set of synonyms and near synonyms, such as “nag”, “steed”, “mare”, “stallion”, “colt”, “pony”, etc., this methodological stance requires that all information they share is to be specified just once, at the conceptual level. The observation that almost all of the specifications, whether considered as “definitory” or encyclopedic, needed to understand one of these terms is relevant to understanding of the others (for instance, the typical shape, sounds emitted, diet, history, behavior, image schemas, etc.) provides an empirical argument in favor of the model this methodologcal stance motivates. Surely the conceptualizer represents all this information about horses, and the parallel sets of idiosyncratic specifications for each of thousands of other cases, just once, at the conceptual level, and specifies at the lexical level for each of these lexemes (“nag”, “colt” etc.) only what is needed to distinguish it from the others. Since each lexeme must to activate this general information store, which we can provisionally call “the concept HORSE”, each lexeme needs a pointer to HORSE. For the lexeme “horse”, that pointer will, arguably, be the whole semantic specification. For “mare”, a specification that its denotation is restricted to females is needed, for “colt” a specification is needed about age, for “pony”, “palomino” clearly something more complex is required (see below). This kind of model is compatible with different pictures of how the conceptual level is structured. For instance, whether there is a theoretically relevant distinction between knowledge of a concept (for instance specifications that enable us to recognize typical horses) and knowledge about the denotation of the concept (e. g. specifications that indicate (correctly or not) that horses were domesticated by Indo-Europeans about five or six thousand years ago somewhere in central Asia, that horses are often raced, that cowboys really love their horses, etc).
Here are a few claims that this model suggests:
(a) One important source of analyticity judgments is the lexical specification. Most conceptual level specifications are not judged to be necessary for the use of a word. For instance, it is not necessary that a horse have the typical horse shape, or size, or color, or neigh, that it eat hay, behave in horsy ways etc. In fact, for some speakers, there are no analytic truths about horses. For others, there are. For instance, some would say that horses are necessarily animals, or even mammals. And, even for those who are willing to imagine that their belief that horses are animals might be wrong, in fact they surely believe very strongly that they are animals, are mammals, have hearts, etc. I think the most striking fact about our strongest beliefs about horses is that their source has nothing specific to do with horses: we have the same set of firm beliefs about living things in general, and the only psychologically plausible model explaining them is one which attributes to adults a rich, and in fact generative, conceptual level structure, a “theory” of living things. Whether this structure gives rise to judgments of necessity seems to depend on how much the judger believes is, how he acquired it, and other factors. For me, and I suppose many others, the strongest source, other than the lexicon, of judgments of necessity is my innately specified ontology. I will return to this below.
But consider the judgments that mares are female, that nags are bad or less than ideal in some vague sense, that colts are young, etc. These seem as sure as any judgments we have about horses, and surer than the judgments listed above (shape, diet, etc.). A plausible explanation for the necessary character of these judgments is that they are trivial matters of lexical specification: they are as close as we can get to something that is “true by definition”, fr words that have no definitions. We do not question them because to question them would be to question our knowledge of the word, and, for these words, we are quite sure we know “what they mean”, i. e., we are sure we know how we use them. My proposal is, then, that the minimal specifications we learn as the lexical specifications of individual words have this character of trivial necessity, whereas the specifications we learn, in various ways, for the concept HORSE do not, in general.
To generalize this as a methodological point: if one has a clear judgment of necessity associated with some word, its source must either be the lexicon or some very deep, perhaps innately specified representational resource, or a mix of both.