Semantics of Noun Compounds
Where is the meaning in Noun Compounds?
In case of binary noun compounds, the core meaning of the compound is obtained from one word known as head word and is modified by another word called modifier. In English, the head of the compound is located on the right. For e.g. School kid is a school going kid, Leather shoes are shoes made of leather, cricket ball is a ball used for playing cricket, etc.
There is another class of Noun Compounds, which get a part of their meaning from the constituent words but gain a special meaning after combination of the constituent words. For e.g. White House. White House is not just any house that is white in color, but refers to the parliamentary house of US president. Similarly, although Green House is a place with green glass walls, it cannot refer to just any green colored house. They are inevitably more difficult to handle.
As we saw, the location of the head of the compound is generally fixed in a language. English compound words are right-headed. Roughly, the head word determines the core property of the compound word and the word at its left acts more or less like a modifier. This modifier can modify any specific property of the head word of the compound. Since we are talking about noun compounds, these properties can be anything like size, color, shape, make, origin, etc.
Let us take the example of the noun “coffee” and the compound words it participates in.
- Black Coffee
- Large Coffee
- Strong Coffee
- Columbian Coffee
We can also take an example of the noun word “table”, from which we get the following noun compound words:
- Round table
- Wooden table
- Dining Table
Interpretation of Noun Compounds using WordNet Similarity
Using WordNet, it is possible to measure semantic similarity between concepts represented by head word and modifier in a Noun Compound. We can take training and test data from annotated corpus like Penn Tree Bank, containing Noun Compounds, and use it to interpret semantic relationship between head and modifier in novel Noun Compounds.
Now, if we have a training data with noun compounds such as “strawberry milk”, “red chair” and “morning walk”, marked with correct relationships of material, color and time respectively (strawberry milk is milk containing strawberry flavor, red is the color of chair and morning walk is a walk done during morning time), we can predict correct semantic relationship between similar noun compounds like “Vanilla shake”, “black table” and “afternoon nap”, with accuracy as Strawberry and Vanilla, black and red as well as “morning” and “afternoon” (modifiers) and “milk” and “shake” (food), “chair” and “table” (furniture items) as well as “walk” and “nap” (activities) will have similar word groupings. Hence, a simple product of individual similarities of each head word and modifier in a Noun Compound can be used to give similar semantic relationship between the given Noun Compound pairing.
© 2017 Payal Khullar. All Rights Reserved.