File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/metho/06/p06-1082_metho.xml
Size: 18,292 bytes
Last Modified: 2025-10-06 14:10:18
<?xml version="1.0" standalone="yes"?> <Paper uid="P06-1082"> <Title>Sydney, July 2006. c(c)2006 Association for Computational Linguistics Word Alignment in English-Hindi Parallel Corpus Using Recency-Vector Approach: Some Studies</Title> <Section position="4" start_page="649" end_page="651" type="metho"> <SectionTitle> 2 Word Alignment Algorithm: Recency </SectionTitle> <Paragraph position="0"/> <Section position="1" start_page="649" end_page="649" type="sub_section"> <SectionTitle> Vector Based Approach </SectionTitle> <Paragraph position="0"> DK-vec algorithm given in (Fung and McKeown, 1994) uses the following dynamic programming based approach to compute the matching cost C(n,m) of two vectors v1 and v2 of lengths n and m, respectively. The cost is calculated recursively using the following formula,</Paragraph> <Paragraph position="2"> where i and j have values from 2 to n and 2 to m respectively, n and m being the number of distinct words in source and target language corpus respectively. Note that vl(k) denotes the kth entry of the vector vl, for l = 1 and 2. The costs are initialised as follows.</Paragraph> <Paragraph position="4"> The word in the target language that has the minimum normalized cost (C(n,m)/(n + m)) is taken as the translation of the word considered in the source text.</Paragraph> <Paragraph position="5"> One major shortcoming of the above scheme is its high computational complexity i.e. O(mn). A variation of the above scheme has been proposed in (Somers, 1998) which has a much lower computational complexity O(min(m,n)). In this new scheme, a distance called Levenshtein distance(S) is successively measured using :</Paragraph> <Paragraph position="7"> The word in the target text having the minimum value of S (Levenshtein difference) is considered to be the translation of the word in the source text.</Paragraph> </Section> <Section position="2" start_page="649" end_page="649" type="sub_section"> <SectionTitle> 2.1 Constraints Used in the Dynamic Programming Algorithms </SectionTitle> <Paragraph position="0"> In order to reduce the complexity of the dynamic programming algorithm certain constraints have been proposed in (Fung and McKeown, 1994).</Paragraph> <Paragraph position="1"> 1. Starting Point Constraint: The constraint imposed is: |first-occurrence of source language word (w1) - first-occurrence of target language word w2 |< 12[?](length of the text). 2. Euclidean distance constraint: The constraint imposed is:radicalbig (m1 [?]m2)2 +(s1 [?]s2)2 < T, where mj and sj are the mean and standard deviation, respectively, of the recency vector of wj, j = 1 or 2. Here, T is some predefined threshold: 3. Length Constraint: The constraint imposed is: 12 [?]f2 < f1 < 2[?]f2, where f1 and f2 are the frequencies of occurrence of w1 and w2, in their respective texts.</Paragraph> </Section> <Section position="3" start_page="649" end_page="649" type="sub_section"> <SectionTitle> 2.2 Experiments with DK-vec Algorithm </SectionTitle> <Paragraph position="0"> The results of the application of this algorithm have been very poor when applied on the three English to Hindi parallel corpora mentioned above without imposing any constraints.</Paragraph> <Paragraph position="1"> We then experimented by varying the values of the parameters in the constraints in order to observe their effects on the accuracy of alignment. As was suggested in (Somers, 1998), we also observed that the Euclidean distance constraint is not very beneficial when the corpus size is small.</Paragraph> <Paragraph position="2"> So this constraint has not been considered in our subsequent experiments. Starting point constraint imposes a range within which the search for the matching word is restricted. Although Fung and McKeown suggested the range to be half of the length of the text, we felt that the optimum value of this range will vary from text to text depending on the type of corpus, length ratio of the two texts etc. Table 2 shows the results obtained on applying the DK vec algorithm on Sentence corpus for different lower values of range. Similar results were obtained for the other two corpora.</Paragraph> <Paragraph position="3"> The maximum increase observed in the F-score is around 0.062 for the Sentence corpus, 0.03 for the Story book corpus and 0.05 for the Advertisement corpus. None of these improvements can be considered to be significant.</Paragraph> </Section> <Section position="4" start_page="649" end_page="650" type="sub_section"> <SectionTitle> 2.3 Experiments with Somers' Algorithm </SectionTitle> <Paragraph position="0"> The algorithm provided by Somers works by first finding all the minimum score word pairs using dynamic programming, and then applying three filters Multiple Alignment Selection filter, Best Alignment Score Selection filter and Frequency Range constraint to the raw results to increase the accuracy of alignment.</Paragraph> <Paragraph position="1"> The Multiple Alignment Selection(MAS) filter takes care of situations where a single target language word is aligned with the number of source language words. Somers has suggested that in such cases only the word pair that has the minimum alignment score should be considered. Table 3 provides results (see column F-score old) when the raw output is passed through the MAS filters for the three corpora. Note that for all the three corpora a variety of frequency ranges have been considered, and we have observed that the results obtained are slightly better when the MAS filter has been used.</Paragraph> <Paragraph position="2"> The best F-score is obtained when frequency range is high i.e. 100-150, 100-200. But here the words are very few in number and are primarily pronoun, determiner or conjunction which are not significant from alignment perspective. Also, it was observed that when medium frequency ranges, such as 30-50, are used the best result, in terms of precision, is around 20-28% for the three corpora. However, since the corpus size is small, here too the available and proposed aligned word pairs are very few (below 25). Lower frequency ranges (viz. 2-20 and its sub-ranges) result in the highest number of aligned pairs. We noticd that these aligned word pairs are typically verb, adjective, noun and adverb. But here too the performance of the algorithm may be considered to be unsatisfactory. Although Somers has recommended words in the frequency ranges 1030 to be considered for alignment, we have considered lower frequency words too in our experiments. This is because the corpus size being small we would otherwise have effectively overlooked many small-frequency words (e.g. noun, verb, adjective) that are significant from the alignment point of view.</Paragraph> <Paragraph position="3"> Somers has further observed that if the Best Alignment Score Selection (BASS) filter is applied to yield the first few best results of alignment the overall quality of the result improves. Figure 1 shows the results of the experiments done for different alignment score cut-off without considering the Frequency Range constraint on the three corpora. However, it was observed that the performance of the algorithm reduced slightly on introducing this BASS filter.</Paragraph> <Paragraph position="4"> The above experiments suggest that the performance of the two algorithms is rather poor in the context of English-Hindi parallel texts as compared to other language pairs as shown by Fung and Somers. In the following section we discuss the reasons for the low recall and precision values.</Paragraph> </Section> <Section position="5" start_page="650" end_page="651" type="sub_section"> <SectionTitle> 2.4 Why Recall and Precision are Low </SectionTitle> <Paragraph position="0"> We observed that the primary reason for the poor performance of the above algorithms in English - Hindi context is the presence of multiple Hindi equivalents for the same English word. This can happen primarily due to three reasons: Declension of Adjective: Declensions of adjectives are not present in English grammar. No morphological variation in adjectives takes place along with the number and gender of the noun. But, in Hindi, adjectives may have such declensions.</Paragraph> <Paragraph position="1"> For example, the Hindi for &quot;black&quot; is kaalaa when the noun is masculine singular number (e.g. black horse [?] kaalaa ghodaa). But the Hindi translation of &quot;black horses&quot; is kaale ghode; whereas &quot;black mare&quot; is translated as kaalii ghodii. Thus the same English word &quot;black&quot; may have three Hindi equivalents kaalaa, kaalii, and kale which are to be used judiciously by considering the number and gender of the noun concerned.</Paragraph> <Paragraph position="2"> Declensions of Pronouns and Nouns: Nouns or pronouns may also have different declensions depending upon the case endings and/or the gender and number of the object. For example, the same English word &quot;my&quot; may have different forms (e.g. meraa, merii, mere) when translated in Hindi.</Paragraph> <Paragraph position="3"> For illustration, while &quot;my book&quot; is translated as [?] merii kitaab, the translation of &quot;my name&quot; is meraa naam. This happens because naam is masculine in Hindi, while kitaab is feminine. (Note that in Hindi there is no concept of Neuter gender). Similar declensions may be found with respect to nouns too. For example, the Hindi equivalent of the word &quot;hour&quot; is ghantaa. In plural form it becomes ghante (e.g. &quot;two hours&quot; [?] do ghante). But when used in a prepositional phrase, it becomes ghanto. Thus the Hindi translation for &quot;in two hours&quot; is do ghanto mein.</Paragraph> <Paragraph position="4"> Verb Morphology: Morphology of verbs in Hindi depends upon the gender, number and per-son of the subject. There are 11 possible suffixes (e.g taa, te, tii, egaa) in Hindi that may be attached to the root Verb to render morphological variations. For illustration, I read. - main padtaa hoon (Masculine) but main padtii hoon (Feminine) You read. - tum padte ho (Masculine) or tum padtii ho (Feminine) He will read. - wah padegaa.</Paragraph> <Paragraph position="5"> Due to the presence of multiple Hindi equivalents, the frequencies of word occurrences differ significantly, and thereby jeopardize the calculations. As a consequence, many English words are wrongly aligned.</Paragraph> <Paragraph position="6"> In the following section we describe certain measures that we propose for improving the efficiency of the recency vector based algorithms for word alignment in English - Hindi parallel texts.</Paragraph> </Section> </Section> <Section position="5" start_page="651" end_page="653" type="metho"> <SectionTitle> 3 Improvements in Word Alignment </SectionTitle> <Paragraph position="0"> In order to take care of morphological variations, we propose to use root words instead of various declensions of the word. For the present work this has been done manually for Hindi. However, algorithms similar to Porter's algorithm may be developed for Hindi too for cleaning a Hindi text of morphological inflections (Ramanathan and Rao, 2003). The modified text, for both English and Hindi, are then subjected to word alignment.</Paragraph> <Paragraph position="1"> Table 4 gives the details about the root word corpus used to improve the result of word alignment. Here the total number of words for the three types of corpora is greater than the total number of words in the original corpus (Table 1). This is because of the presence of words like &quot;I'll&quot; in the English corpus which have been taken as &quot;I shall&quot; in the root word corpus. Also words like Unkaa have been taken as Un kaa in the Hindi root word corpus, leading to an increase in the corpus size.</Paragraph> <Paragraph position="2"> Since we observed (see Section 2.2) that Euclidean distance constraint does not add significantly to the performance, we propose not to use this constraint for English-Hindi word alignment.</Paragraph> <Paragraph position="3"> However, we propose to impose both frequency range constraint and length constraint (see Section 2.1 and Section 2.3). Instead of the starting point constraint, we have introduced a new constraint, viz. segment constraint, to localise the search for the matching words. The starting point constraint expresses range in terms of number of words. However, it has been observed (see section 2.2) that the optimum value of the range varies with the nature of text. Hence no value for range may be identified that applies uniformly on different corpora. Also for noisy corpora the segment constraint is expected to yield better results as the search here is localised better. The proposed segment constraint expresses range in terms of segments. In order to impose this constraint, first the parallel texts are aligned at sentence level. The search for a target language word is then restricted to few segments above and below the current one.</Paragraph> <Paragraph position="4"> Use of sententially aligned corpora for word alignment has already been recommended in (Brown et al., 1993). However, the requirement there is quite stringent - all the sentences are to be correctly aligned. The segment constraint proposed herein works well even if the text alignment is not perfect. Use of roughly aligned corpora has also been proposed in (Dagan and Gale, 1993) for word alignment in bilingual corpora, where statistical techniques have been used as the underlying alignment scheme. In this work, the sentence level alignment algorithm given in (Gale and Church, 1991) has been used for applying segment constraint. As shown in Table 5, the alignment obtained using this algorithm is not very good (only 70% precision for Storybook corpus). The three aligned root word corpora are then subjected to segment constraint in our experiments.</Paragraph> <Paragraph position="5"> Next important decision we need to take which dynamic programming algorithm should be used.</Paragraph> <Paragraph position="6"> Results shown in Section 2.2 and 2.3 demonstrate that the performance of DK-vec algorithm and Somers' algorithm are almost at par. Hence keeping in view the improved computational complexity, we choose to use Levenshtein distance as used in Somers' algorithm for comparing recency vectors. In the following subsection we discuss the experimental results of the proposed approach.</Paragraph> <Section position="1" start_page="652" end_page="653" type="sub_section"> <SectionTitle> 3.1 Experimental Results and Comparison </SectionTitle> <Paragraph position="0"> with Existing algorithms We have conducted experiments to determine the number of segments above and below the current segment that should be considered for searching the match of a word for each corpus. In this respect we define i-segment constraint in which the search is restricted to the segments k [?]i to k + i of the target language corpus when the word under consideration is in the segment k of the source language corpus.</Paragraph> <Paragraph position="1"> Evidently, the value of i depends on the accuracy of sentence alignment. Table 5 suggests that the quality of alignment is different for the three corpora that we considered. Due to the very high precision and recall for Sentence corpus we have restricted our search to the kth segment only, i.e. the value of i is 0. However, since the results are not so good for the Storybook and Advertisement corpora we found after experimenting that the best results were obtained when i was 1. During the experiments it was observed that as the number of segments was lowered or increased from the optimum segment the accuracy of alignment decreased continuously by around 10% for low frequency ranges for the three corpora and remained almost same for high frequency ranges.</Paragraph> <Paragraph position="2"> Table 3 shows the results obtained when segment constraint is applied on the three corpora at optimum segment range for various frequency ranges. A comparison between the F-score given by algorithm in (Somers, 1998) (the column F-score old in the table) and the F-score obtained by applying the improved scheme (the column F-score new in the table) indicate that the results have improved significantly for low frequency ranges.</Paragraph> <Paragraph position="3"> It is observed that the accuracy of alignment for almost 95% of the available words has increased significantly. This accounts for words within low frequency range of 2-40 for Sentence corpus, 230 for Storybook corpus, and 2-20 for Advertisement corpus. Also, most of the correct word pairs given by the modified approach are verbs, adjectives or nouns. Also it was observed that as the noise in the corpus increased the results became poorer. This accounts for the lowest F-score values for advertisement corpus. The Sentence corpus, however, has been found to be the least noisy, and highest precision and recall values were obtained with this corpus.</Paragraph> <Paragraph position="4"> Using Somers' second filter on each corpus for the optimum segment we found that the results at low scores were better as shown in Figure 1. The word pairs obtained after applying the modified approach can be used as anchor points for further alignment as well as for vocabulary extraction. In case of the Sentence corpus, best result for anchor points for further alignment lies at the score cut off 1000 where precision and recall are 86.88% and 80.35% respectively. Hence F-score is 0.835 which is very high as compared to 0.173 obtained by Somers' approach and indicates an improvement of 382.65%. Also, here the number of correct word pairs is 198, whereas the algorithms in (Fung and McKeown, 1994) and (Somers, 1998) gave only 62 and 61 correct word pairs, respectively. Hence the results are very useful for vocabulary extraction as well. Similarly, Figure 2 and Figure 3 show significant improvements for the other two corpora. At any score cut-off, the modified approach gives better results than the algorithms proposed in (Somers, 1998).</Paragraph> </Section> </Section> class="xml-element"></Paper>