ABSTRACT
At the present time, there is a resurgence of MT research based only on text statistics at IBM, New York, a revived technique (Brown et al. 1990) that has attracted both interest and funding. Their claim that MT can and should be done on the basis of correlations of English and foreign words, established between very large bilingual corpora, assumes that ‘symbolic,’ non-quantitative, MT cannot do the job. The quick answer to them is again SYSTRAN, which IB M’s ‘proportion of sentences correct’ percentage (40 per cent versus SYSTRAN’S 60-70 percent success rate) lags far behind, with no evidence beyond hope, energy and application of ever closing the gap. IBM’s argument simply ignores what constitutes a refutation of their claims, namely, SYSTRAN.
