Word compounding is a morphological phenomenon where two or more base
lexemes join together by concatenation to form a new lexeme, compounding is widely
used in many languages such as English, French and it is particularly common in
modern Chinese. Since the rules for the compounding words are quite productive, the
semantics relationship between base lexemes and its compounded word is not always
easy to interpret, especially for an automated system. For example, many internet
slangs are created by compounding. A frequency-based word embedding model will
have difficulty to deal with these out-of-vocabulary words.
That is the reason why understanding these relationships could be useful for
overcoming this limit.

Word compounding is a morphological phenomenon where two or more base
lexemes join together by concatenation to form a new lexeme, compounding is widely
used in many languages such as English, French and it is particularly common in
modern Chinese. Since the rules for the compounding words are quite productive, the
semantics relationship between base lexemes and its compounded word is not always
easy to interpret, especially for an automated system. For example, many internet
slangs are created by compounding. A frequency-based word embedding model will
have difficulty to deal with these out-of-vocabulary words.
That is the reason why understanding these relationships could be useful for
overcoming this limit.
There are four types of compound words:

1. Endocentric: one lexeme serves as a ‘selector’ of the other lexeme. The
   compounded word denotes a subset of one lexeme.in this case, the selected
   member usually bears the semantic head of the compound. In terms of word
   order: the selectors are usually the first lexeme, and the semantic heads are
   the second one.
   E.g. ‘Microclimate’ is the climate of a very small or restricted area
2. Exocentric: the meaning of the compound can be denoted by neither its
   components. The semantic head cannot be found in the compound.
   E.g. ‘Sweetheart’ is not something sweet, nor something about the heart.
3. Appositional: two lexemes of the compound refer to the same entity.
   E.g. ‘actor-director’: someone is an actor and also a director.
4. Coordinative: two lexemes of the compound are both semantic heads

In this paper, they proposed a general framework for representing the meaning of
phrase and sentences in vector space, more precisely, they formulated semantic
composition of two words as a function with two variables and two free parameters,
the general form of this function is :
𝑞 = 𝑓(𝑣,𝑤,𝑅,𝐾)
Where f is a function (linear or non-linear) from 𝑅 𝑜 to 𝑅 𝑛,n is not necessarily equal to m.

u and v are vector representations (of dimension n) of the component words, R
represents some syntactic relation, K represents any additional information, and
finally, p is the result of the composition process.