Hybrid Approach to Zero Subject Resolution for multilingual MT - Spanish-to-KoreanCases -

19227

Authors

nan

Abstract

The current paper proposes a novel approach to Spanish zero pronoun resolution in the context of Spanish to Korean Machine Translation (MT). Spanish is one of the well-known 'pro-drop' languages so that especially a subject pronoun is often omitted, if it can be inferred from the linguistic as well as non- linguistic context. In Spanish to Korean MT the omitted subject doesn't need to be restored in many cases as Korean also allows a zero subject. However, there are some cases where the omitted subject must be identified to ensure a correct translation. To restore the omitted subject, linguistic clues can be employed, as Spanish verbs undergo morphological flections with respect to the gender and number. However, there still remain some ambiguous cases in which there are more than two possible subject candidates for the specific verb endings. In this paper, we propose a hybrid approach to resolve Spanish zero subject that integrates linguistic knowledge (morphological information) and artificial intelligence knowledge (machine learning approach). We proposed 11 linguistically motivated features for ML (Machine Learning). Our approach has been implemented with WEKA 3.6.10 and evaluated by using 10 fold cross validation method. The accuracy of the proposed method reached 83.6% while the baseline method that randomly chooses a possible subject candidate among three most frequent subject types shows only 33.3% accuracy rate. 1

Keywords

nan

Full Content

Hybrid Approach to Zero Subject Resolution for multilingual MT - Spanish-to-Korean Cases - Arum Park Munpyo Hong* Dept. of German Linguistics & Literature, Sungkyunkwan University / 25-2, Sungkyunkwan-Ro, Jongno-Gu, Seoul, Korea Dept. of German Linguistics & Literature, Sungkyunkwan University / 25-2, Sungkyunkwan-Ro, Jongno-Gu, Seoul, Korea remin2@skku.edu skkhmp@skku.edu Abstract The current paper proposes a novel approach to Spanish zero pronoun resolution in the context of Spanish to Korean Machine Translation (MT). Spanish is one of the well-known 'pro-drop' languages so that especially a subject pronoun is often omitted, if it can be inferred from the linguistic as well as non- linguistic context. In Spanish to Korean MT the omitted subject doesn't need to be restored in many cases as Korean also allows a zero subject. However, there are some cases where the omitted subject must be identified to ensure a correct translation. To restore the omitted subject, linguistic clues can be employed, as Spanish verbs undergo morphological flections with respect to the gender and number. However, there still remain some ambiguous cases in which there are more than two possible subject candidates for the specific verb endings. In this paper, we propose a hybrid approach to resolve Spanish zero subject that integrates linguistic knowledge (morphological information) and artificial intelligence knowledge (machine learning approach). We proposed 11 linguistically motivated features for ML (Machine Learning). Our approach has been implemented with WEKA 3.6.10 and evaluated by using 10 fold cross validation method. The accuracy of the proposed method reached 83.6% while the baseline method that randomly chooses a possible subject candidate among three most frequent subject types shows only 33.3% accuracy rate. 1 Introduction Spanish is one of the so-called pro-drop languages where certain pronouns may be omitted. In Spanish, the pronominal subject can be deleted and is called a zero subject. A zero subject is the most frequent type of anaphoric expression in Spanish.1 Spanish zero subject is one of the important issues that must be tackled in Spanish-to-Korean MT. This kind of pronoun is very important due to its high frequency in Spanish texts. In many cases its resolution is obligatory in Spanish-to-Korean MT. Palomar et al.(2001) reported that about 65.5% of the pronouns are the zero subject pronoun in pronoun occurrences in Spanish corpus. Let us consider the following example. In this example, the omitted subject is represented by the symbol ø. (1) Luis quiere que Ø vayamos[1st plural] a la playa. Luis want that go to the beach * Corresponding author 1 Palomar et al.(2001) PACLIC 28 254 Copyright 2014 by Arum Park and Munpyo Hong 28th Pacific Asia Conference on Language, Information and Computation pages 254–261 lwuisunun wulituli haypyeney kakilul wunhanta “Luis wants us to go to the beach.” In Spanish, the subject pronoun and the verb must agree in person and number. Even though the subject pronoun is not present in the sentence, the zero subject can be restored from the verb ending, as '-os' is a 1st person plural morpheme. This gives us clues to resolve …

Content truncated to 500 words

Labeled Text

The current paper proposes a novel approach to Spanish zero pronoun resolution in the context of Spanish to Korean Machine Translation (MT). Spanish is one of the well-known 'pro-drop' languages so that especially a subject pronoun is often omitted, if it can be inferred from the linguistic as well as nonlinguistic context. In Spanish to Korean MT the omitted subject doesn't need to be restored in many cases as Korean also allows a zero subject. However, there are some cases where the omitted subject must be identified to ensure a correct translation. To restore the omitted subject, linguistic clues can be employed, as Spanish verbs undergo morphological flections with respect to the gender and number. However, there still remain some ambiguous cases in which there are more than two possible subject candidates for the specific verb endings. In this paper, we propose a hybrid approach to resolve Spanish zero subject that integrates linguistic knowledge (morphological information) and artificial intelligence knowledge (machine learning approach). We proposed 11 linguistically motivated features for ML (Machine Learning). Our approach has been implemented with WEKA 3.6.10 and evaluated by using 10 fold cross validation method. The accuracy of the proposed method reached 83.6% while the baseline method that randomly chooses a possible subject candidate among three most frequent subject types shows only 33.3% accuracy rate. 1

Publication Details

Paper ID:
19227
Year:
2014
Publication:
TIPSTER
Publication Type:
conf
DOI:
nan