I promised some further posts on the topic of metaphor, and on the conventionalization of metaphor in particular, but in order to get to that, we need to get some things out of the way first. Let’s start with polysemy. If you don’t know what polysemy is, then you need to study up on your Greek roots. Kidding, of course. Polysemy refers to a single word having multiple related meanings or senses. “Bank” is a good example of a polysemous word, especially since using it will help later in the post. Bank can mean the banking company, like Fifth Third Bank (that name has always cracked me up), it can mean the building where you deposit and withdraw your money or apply for a loan, or it can mean the action of doing bank stuff (“I bank at Fifth Third Bank; you know, the bank on Elm Street”). An important question for psychologists and psycholinguists is, how do we represent the different senses of polysemous words? Are they all represented as a single lexical unit (i.e., as one entry in our mental lexicon), called a lemma by linguists, or are they represented as separate lemmas?
To see the difference between one or multiple lexical units, consider the case of homonyms. Homonyms, as I’m sure you’re all aware, are words that have two or more unrelated meanings. “Bank” is again a good example. There’s the financial institution meaning of bank, and then there’s the incline next to a creek or river. These meanings are unrelated, and the fact that they are referred to using the same phonemes in the same order is probably just a coincidence. It’s pretty clear that the two unrelated meanings of “bank” are represented as separate lemmas. So we can rephrase the question about the representation of polysemous senses like this: are they represented like words with a single sense, are they represented like homonyms, or are their representations some hybrid of the two?
Greg Murphy and his colleagues have been doing really interesting work on this question lately, and they’ve come to the conclusion that polysemous senses are represented separately, like homonymous meanings. Their first line of evidence comes from experiments showing that polysemous senses, like homonymous meanings, interfere with each other. In one experiment1 Klein and Murphy had people learn a set of polysemous words presented in phrases (e.g., “daily paper”). After a delay, they were presented with a bunch of phrases, one at a time, and asked to indicate whether the capitalized word (which would be the polysemous word) had been in the set they’d previously learned. The key manipulation was how the word was presented. There were three conditions: a same phrase condition, in which participants read the word in the same phrase that they had learned it (“daily PAPER”); a different phrase, same sense condition, in which a different modifier was used, but the same sense of the word was retained (e.g., “liberal PAPER,” in which “paper” still refers to a newspaper); and a different sense condition, in which a different but related (i.e., polysemous) meaning of the word was presented (e.g., “wrapping PAPER”). Clearly, using the same phrase should produce the best memory accuracy, so the key comparison is between the different phrase, same sense condition and the different sense condition. If polysemous senses are represented as one lexical unit, then there should be no different between these conditions, but if they are represented as different lexical units, then accuracy in the different sense condition should be lower. That’s what happens when you use homonyms in this paradigm.
The accuracy rates for the same phrase, different phrase/same sense, and different sense conditions were 76%, 64%, and 56%, respectively. Each of those differences was statistically significant. To provide further evidence, Klein and Murphy conducted a second experiment using a different paradigm. When you prime people with one meaning of a homonymous word, and then give them another unrelated meaning of the term in a phrase and ask them to say as quickly as they can whether the phrase makes sense, they’re slower to do so than if you give them the same meaning both in the prime and in the sense/nonsense task. So, if polysemous senses of a word are reprsented separately, like homonmyous meanings, you should see the same pattern. Klein and Murphy found this using phrases like those in the previous experiment: particpants were significantly slower to verify that phrases containing polysemous words made sense if different senses were used in the prime and sense/nonsense task, than if the same senses were used. From these two experiments, it appears that polysemous senses interfere with each other, just as homonymous meanings do.
The next question, then, is how the separate representations of polysemous senses are related to each other. Klein and Murphy, in a seconc paper2, starting with the assumption that words “pick out” categories, offer several possibilities. Polysemous senses could be related to each other taxonomically, as many noun categories are (e.g., ANIMAL, BIRD, FINCH). They could also be related to each other thematically, like, say, the concepts BED and PILLOW. Finally, they might be related in an ad hoc fashion, as the members of the categories “things to take on a plane” and “things to take out of a burning house.”
In order to explore these possibilities, Klein and Murphy first looked at how people would form categories with polysemous senses. Their first three experiments used a forced sorting task. Force sorting tasks present people with one item, and then with two or more items that are related to the first thing in different ways (or are completely unrelated), and ask participants to choose the item that goes best with the first item. In their experiments, participants were presented with a phrase (the “target phrase”) that used a polysemous word. They were then asked to choose between a phrase containing the same word in the same sense or in a different but related sense, or a phrase containing either a thematically or taxonomically related word. For example. particpiants might have been presented with the target phrase “wrapping PAPER.” They would then be given another phrase with the word “pape” in the same sens, like “shredded PAPER”, or a different sense such as “liberal PAPER,” along with either a phrase containing a taxonomically related concept such as “smooth CLOTH” (paper and cloth are both types of material) or a thematically related concept such as “sharp SCISSORS” (scissors cut paper… rocks smash scissors).
In the first two experiments, they found that 80% of the time, participants chose either the taxonomically related or thematically related phrase over the phrase containing a different polysemous sense of the word in the target phrase, but when one choice was the target word used in the same sense, participants chose this 70% of the time. Murphy and Klein take this as implying that people don’t treat polysemous senses as forming a category. In a third experiment, however, they found that participants chose the related sense over the taxonomically or thematically related category 15% of the time, but when the choice was a homonymous meaning of the word in the target phrase or a taxonomically or thematically related word, participants chose the homonymous meaning only 7% of the time. So it appears that while polysemous senses are not very related, they are more related than homonymous meanings. In two follow-up experiments, participants continued to choose the phrase containing the related sense of the target word about 15% of the time even when they were “hit over the head” with the connection between the polysemous senses of the words prior to completing the forced sorting task.
So far, it looks like polysemous senses are related to each other in much the same way that homonymous senses are. However, in their sixth experiment, (11th, if you count the two papers together), they finally found a difference. For both taxonomically and, to a lesser extent, thematically related categories, category membership is a basis for induction. For example, if we know that both robins and finches are members of the category BIRD, and we are told that robins have two spleens, we’re likely to believe that the probability that finches have two spleens is pretty high, whereas knowing that robins have two spleens (they don’t, really, I don’t think) wouldn’t have much influence on whether we believe salmon have two spleens (especially since we know that some members of the category ANIMAL, a category in which both robins and salmon are members, don’t have two spleens). Klein and Murphy’s sixth experiment was designed to test whether people are willing to make inductions from one sense of a polysemous word to another, much as they will make inductions from one member of a category to another. Here’s an example from their task:
Suppose that scientists find the biotin bacteria in wrapping PAPER.
Type in the probability (out of 100) that the biotin bacteria will also be in liberal PAPER.
In addition to polysemous words, they also tested homonymous meanings. They found that the probabilities participants listed for inductions between polysemous senses were 1.5 to 2 (around 40%) higher than those they listed for inductions between homonymous meanings (20-30%).
At this point, you might be wondering what this has to do with dead metaphors, or maybe you’ve put two and two together. A full discussion of that will have to wait for a subsequent post, but for now I will quickly note that if metaphorical senses of a word become conventional, i.e., if they create a new sense for a polysemous word, the new sense will be stored separately from the original sense, and the relationship between the new sense and the sense from which it was derived will be only slightly stronger than that between homonyms.
1Klein, D. E., & Murphy, G. L. (2001). The representation of polysemous words. Journal of Memory and Language, 45, 259-282.
2Klein, D. E., & Murphy, G. L. (2002). Paper has been my ruin: Conceptual relations of polysemous senses. Journal of Memory and Language, 47, 548-570.