Random vector generation of a semantic space

We show how random vectors and random projecon can be implemented in the usual vector space model to construct a Euclidean semanc space from a French synonym diconary. We evaluate theorecally the resulng noise and show the experimental distribuon of the similaries of terms in a neighborhood according to the choice of parameters. We also show that the Schmidt orthogonalizaon process is applicable and can be used to separate homonyms with disnct semanc meanings. Neighboring terms are easily arranged into semancally significant clusters which are well suited to the generaon of realisc lists of synonyms and to such applicaons as word selecon for automac text generaon. This process, applicable to any language, can easily be extended to collocaons, is extremely fast and can be updated in real me, whenever new synonyms are proposed. 1. Introducon In their seminal work, Ploux and Victorri 1 have used synonymy relaons deduced from French electronic diconaries to create semanc spaces around French words and their neighbors. Their definion of " synonymy " is fairly broad and includes hyponymy (moineau and oiseau), hyperonymy (arme and pistolet) or even non‐synonymous, but related terms (autocar and automobile); however, in their work, true synonyms (i.e. terms which are more or less interchangeable) form cliques of the graph of synonyms, i.e. maximally complete subgraphs. While this is very interesng from a theorecal standpoint, as it then becomes straigh跀í°€orward to evaluate an interclique distance (or degree of separaon) between any two terms in the graph (as long as neither belongs to an island, such as lapereau and lapinot), it is not very useful in pracce. For example, an author in search of the right term may well not be interested in strict synonyms; terms with related or even opposed meanings can oen be preferable in rhetorical figures. Also, in many applicaons such as automac text generaon, a well‐defined and mathemacally well behaved semanc distance between terms is oen a prerequisite. In this report, we show how an Euclidean semanc distance can quickly and easily be constructed from Ploux and Victorri's database (which contains 54,685 terms and 116,694 cliques). Since the pioneering work of Salton 2 , 3 , it is well understood that any combinaon of terms, such as a clique, can be seen as a vector in a space where each dimension represents a disnct term (or lemma.) (1) This representaon is extremely frui跀í°€ul and forms the basis of …