Aljlayl, M.; Frieder, O.; Grossman, D.: On bidirectional English-Arabic search (2002)
0.00
0.0033821356 = product of:
0.0067642713 = sum of:
0.0067642713 = product of:
0.020292813 = sum of:
0.020292813 = weight(_text_:12 in 5227) [ClassicSimilarity], result of:
0.020292813 = score(doc=5227,freq=2.0), product of:
0.13281173 = queryWeight, product of:
2.765864 = idf(docFreq=7562, maxDocs=44218)
0.048018172 = queryNorm
0.15279384 = fieldWeight in 5227, product of:
1.4142135 = tf(freq=2.0), with freq of:
2.0 = termFreq=2.0
2.765864 = idf(docFreq=7562, maxDocs=44218)
0.0390625 = fieldNorm(doc=5227)
0.33333334 = coord(1/3)
0.5 = coord(1/2)
- Abstract
- Aljlayl, Frieder, and Grossman review machine translation of query methodologies and apply them to English-Arabic/Arabic-English Cross-Language Information Retrieval. In the dictionary method, replacement of each term with all possible equivalents in the target language results in considerable ambiguity, while taking the first term in the dictionary list reduces the ambiguity but may fail to capture the meaning. A Two-Phase method takes all possible equivalents and translates them back, retaining only those that generate the original term. It results in an average query length of six terms in TREC7 and 12 in TREC9. Arabic to English translations consistently preformed below the original English queries, and the Two-Phase method consistently preformed at the highest level and significantly better than the Every-Match method. Machine translation using other techniques is economical for queries but not likely so for documents. Using ALKAFI, a commercial translation system from Arabic to English and the Al-Mutarjim Al-Arabey system for English to Arabic, nearly 60% of monolingual retrievals were generated going from Arabic to English. Smaller numbers of terms in the source query improve performance, and these systems require syntactically well-formed queries for good performance.
