Lexical decomposition is the belief that word meanings can be built up compositionally from the meanings of simpler words. In the same way that every number can be expressed as the sum of several primes, each word in the lexicon would be the result of combining the senses of several semantic primes. A word could then be decomposed to reveal its semantic ancestry.

For example: Stallion = Horse + Male

In this case, 'stallion' is built up from the more primitive concepts 'horse' and 'male'. The motivation for pursuing lexical decomposition as a means of understanding word sense has historically been based on the fact that no matter how many different sound combinations we find in a language, it is possible to trace them all back to a finite and usually small set of phonemes. In English, for instance, there are approximately 30-50 phonemes which combine to produce every single word in the English language. If we can produce the sound of every word from such a small set, theoretically it should be possible to produce the meaning of every word from a similar set of basic concepts.

One of the reasons people believe that lexical decomposition is valid is because many words contain partial similarities. A mare is a female horse whereas a stallion is a male horse. They both contain a vital element and then vary slightly, almost as if they contain different ingredients. Similarly, if we mix these ingredients, we can find correlations across many concepts.

       Male     Female
Sheep  ram       ewe
Horse  stallion  mare
Cow    bull      cow
It is important to note that all of this is based on the assumption that there is some amount of discreteness in the meanings, that one can pick apart 'stallion' to mean 'male' and 'horse', that these senses somehow still can be found within the word. While this may seem to be a rather artificial notion of word sense, there is a lot of evidence that shows that such a discreteness does in fact exist:
  • "I almost killed her" is an ambiguous sentence which could be interpreted as either that the speaker was about to engage in killing the woman, or that he had harmed her so severely that she was almost dead. This implies that kill is made up of two discrete senses, cause and die. If this is true then the first interpretation can be constructed by augmenting cause with "almost" (almost caused her to die), and the second could be formed by augmenting die with "almost" (caused her to almost die).

  • "The astronaut re-entered the atmosphere" is in some ways a strange sentence because re-enter works even if it is the first time that the astronaut is returning to the atmosphere (re-enter would suggest that he entered it previously from space, and is now entering it once again). This suggests that in this case 'enter' means move + in, and "re-" only applies to in (move in the atmosphere again).

  • "That's not a stallion" typically causes one to think that the animal is in fact a mare, suggesting that 'stallion' is horse + male and only male is being negated. (Similar situation if we say the phrase "That's not a man!". We assume then that the person is a woman) In the cases where the full sentence would be "That's not a stallion, that's a ewe!", both components of 'stallion' could be negated.

Problems with Lexical Decomposition

One of the main problems with lexical decomposition is that many times our decomposition leads us to bogus analyses. Let's take the example we used before:

Kill = Cause + Die

This decomposition worked very well in explaining the ambiguousness of "almost killed", but we are left with the problem that in many cases, 'kill' does not actually equal the sum of the senses of 'cause' and 'die'.

1. John caused Bill to die on Saturday by poisoning his soup on Friday.
2. John killed Bill on Saturday by poisoning his soup on Friday.

The first sentence is correct, while the second seems a bit odd. There are senses where 'cause' and 'die' seem to make sense, but differ from the meaning of 'kill'. The response to this is that these meanings are very abstract and should not be equated completely the meanings of the real words. When we say that Kill = Cause + Die, we don't mean that it has the concrete meanings of 'cause' and 'die', merely that it means generally what they mean abstractly. The linguist Anna Wierzbicka has argued against this interpretation, arguing that these meanings should in fact be very concrete and intuitable. She criticizes this method of dealing with the abstract meanings, calling it more of a translation into an artifical language than a real analysis of word sense.

Finally, one of the other objections is that if we say that Stallion = Horse + Male and take either one of the components away, the other one still is intelligible, but if Horse = Animal + Equine (as it is commonly held to be) and take away Animal, we are left with Equine, but what does Equine mean in the absence of Animal? The only way to explain it is to relate it to Horse, perhaps like Equine = "pertaining to horses". If this is true then, we are left with the fact that Horse is a "horsey animal". Is this a real analysis of the meaning of the word? Even if we do consider it to be a valid and meaningful analysis, it is problematic because it seems to be of a rather different nature than Stallion = Horse + Male. Is this acceptable?


Information from:
Cruse, Alan, Meaning in Language: An Introduction to Semantics and Pragmatics, Oxford University Press, Oxford, 2000.