Lower and higher estimates of the number of “true analogies” between sentences contained in a large multilingual corpus

Lower and higher estimates of the number of “true analogies” between sentences contained in a large multilingual corpus Yves Lepage ATR Spoken Language Translation Research Labs Hikari-dai 2-2-2, Keihanna Science City 619-0288 Ky¯oto Japan yves.lepage@atr.jp Abstract The reality of analogies between words is re- futed by noone (e.g., I walked is to to walk as I laughed is to to laugh, noted I walked : to walk :: I laughed : to laugh). But computational lin- guists seem to be quite dubious about analogies between sentences: they would not be enough numerous to be of any use. We report experi- ments conducted on a multilingual corpus to es- timate the number of analogies among the sen- tences that it contains. We give two estimates, a lower one and a higher one. As an analogy must be valid on the level of form as well as on the level of meaning, we relied on the idea that translation should preserve meaning to test for similar meanings. 1 Introduction A long tradition in linguistics views analogy as a means for the speaker to analyze or produce new sentences1. To be linguistically relevant, an analogy should hold on the level of form as well as on the level of meaning. In contrast to that, in Greek and Latin antiquity, anomaly designated those cases2 where an analogy of meaning is not reflected by an analogy of form. (e.g., ‘I drink.’ : ‘I’d like to drink.’ :: ‘I can swim.’ : ‘I’d like to be able to swim.’3). Conversely, the existence of analogies of form that are not verified on the level of meaning has been taken by the Generativists to indicate the indepen- dence of syntax (e.g., Abby is baking vegan pies. : Abby is baking. :: Abby is too tasteful to pour gravy 1See, inter alia, (PAUL, 1920, chap.5), (de SAUSSURE, 1995, 3rd part, chap. iv), (BLOOMFIELD, 1933, p.276), (MOUNIN, 1968, p.119–120), (ITKONEN, 1994, p.48–50), (PULLUM, 1999, p.340–343). 2In those times, the cases considered were in fact in mor- phology (See, .e.g., VARRO, De lingua latine). 3The meaning ‘I’d like to be able to swim.’ cannot be con- strued as ∗I’d like to can swim. on vegan pies. : Abby is too tasteful to pour gravy on.4). The purpose of this study is to estimate the num- ber of “true analogies” present in a large corpus, i.e., analogies which hold both on the level of form, as well as on the level of meaning. Formally, let us denote ‘A’ as some meaning, and L(‘A’) as the set of all possible ways of realising ‘A’ in a particular language L. Let us denote A as some realisation of ‘A’, i.e., A ∈ L(‘A’). With these notations, we want to count, in a given corpus, all cases where the following holds5. A ∈ L(‘A’) B ∈ L(‘B’)∧ ∧ C ∈ L(‘C’) D ∈ L(‘D’)∧ ∧ A : B :: C : D ∧ ‘A’ : ‘B’ :: ‘C’ : ‘D’ The reason for estimating the …

Content truncated to 500 words