Translating Compounds by Learning Component Gloss Translation Models via Multiple Languages

Translating Compounds by Learning Component Gloss Translation Models via Multiple Languages Nikesh Garera and David Yarowsky Department of Computer Science Center for Language and Speech Processing Johns Hopkins University Baltimore, MD 21218, USA {ngarera,yarowsky}@cs.jhu.edu Abstract This paper presents an approach to the translation of compound words without the need for bilingual training text, by mod- eling the mapping of literal component word glosses (e.g. “iron-path”) into flu- ent English (e.g. “railway”) across mul- tiple languages. Performance is improved by adding component-sequence and learned- morphology models along with context sim- ilarity from monolingual text and optional combination with traditional bilingual-text- based translation discovery. 1 Introduction Compound words such as lighthouse and fireplace are words that are composed of two or more compo- nent words and are often a challenge for machine translation due to their potentially complex com- pounding behavior and ambiguous interpretations (Rackow et al., 1992). For many languages, such words form a significant portion of the lexicon and the compounding process is further complicated by diverse morphological processes (Levi, 1978) and the properties of different compound sequences such as Noun-Noun, Adj-Adj, Adj-Noun, Verb-Verb, etc. Compounds also tend to have a high type frequency but a low token frequency which makes their transla- tion difficult to learn using corpus-based algorithms (Tanaka and Baldwin, 2003). Furthermore, most of the literature on compound translation has been re- stricted to a few languages dealing with compound- ing phenomena specific to the language in question. Compound Splitting English Gloss Translation Input: Distilled glosses from German-English dictionary Krankenhaus Kranken-Haus sick-house hospital Regenschirm Regen-Schirm rain-guard umbrella W¨orterBuch W¨orter-Buch words-book dictionary Eisenbahn Eisen-Bahn iron-path railroad Input: Distilled glosses from Swedish-English dictionary Sjukhus Sjhu-Khus sick-house hospital J¨arnv¨ag J¨arn-v¨ag iron-path railway Ordbok Ord-Bok words-book dictionary Goal: To translate new Albanian compounds Hekurudh¨e Hekur-Udh¨e iron-path ??? Table 1: Example lexical resources used in this task and their application to translating compound words in new languages. With these challenges in mind, the primary goal of this work is to improve the coverage of translation lexicons for compounds, as illustrated in Table 1 and Figure 1, in multiple new languages. We show how using cross-language compound evidence ob- tained from bilingual dictionaries can aid in com- pound translation. A primary motivating idea for this work is that the literal component glosses for compound words (such as “iron path” for railway) is often replicated in multiple languages, providing insight into the fluent translation of a similar literal 2gloss in a new (often resource-poor) language. Resources Utilized The only resource utilized for our compound trans- lation lexicon algorithm is a collection of bilingual dictionaries. We used bilingual dictionary collec- tions for 50 languages that were acquired in elec- tronic form over the Internet or via optical character recognition (OCR) on paper dictionaries. Note that no parallel or even monolingual corpora is required, their use described later in the paper is optional. 403 3 Related Work The compound-translation literature typically deals with these steps: 1) Compound splitting, 2) transla- tion candidate generation and 3) translation …

Content truncated to 500 words