C SC 620 Advanced Topics in Natural Language ProcessingMachine TranslationPaper 5: A Framework for Syntactic Translation. V. YngveSlide 4Slide 5Slide 6Slide 7Slide 8Slide 9Slide 10Slide 11Slide 12Slide 13Slide 14Slide 15Slide 16Slide 17C SC 620Advanced Topics in Natural Language Processing3/9Lecture 14Machine Translation•Readings in Machine Translation, Eds. Nirenburg, S. et al. MIT Press 2003.•Part 1: Historical Perspective•Reading list:–Introduction. Nirenburg, S.–1. Translation. Weaver, W.–3. The Mechanical Determination of Meaning. Reifer, E.–5. A Framework for Syntactic Translation. Yngve, V.–6. The Present Status of Automatic Translation of Languages. Bar-Hillel, Y.Paper 5: A Framework for Syntactic Translation. V. Yngve•Introduction: MIT Approach–“We are attempting to go beyond simple word-for-word translation: beyond translation using empirical, ad hoc, or pragmatic syntactic routines. The concept of full syntactic translationhas emerged: translation based on a thorough understanding of linguistic structures, their equivalences, and meanings.”Paper 5: A Framework for Syntactic Translation. V. Yngve•Contextual clues:1. The field of discourse•specialized dictionaries2. Recognition of coherent word groups•idioms and compound nouns3. The syntactic function of each word•help solve word order problems as well as multiple-meaning problems•(German) der –article, relative or demonstrative pronoun–aominative, genitive or dativePaper 5: A Framework for Syntactic Translation. V. Yngve•Contextual clues:4. Selectional relations between open class items•nouns, verbs, adjectives and adverbs5. Antecedents•pronouns etc.6. All other contextual cues, especially those concerned with an exact knowledge of the subject under discussion•… last to be mechanisedPaper 5: A Framework for Syntactic Translation. V. Yngve•Two Approaches–95% approach–Full understanding•Decoding verb, noun and adjective inflection–Can do a perfect job: one need not be satisfied with anything less•Verb category translation–? Auxiliary, ergative, unergative–Imperfectly understoodPaper 5: A Framework for Syntactic Translation. V. Yngve•Syntactic Translation–Examination of the 6 types of clues above reveals that they are predominantly concerned with the relationships of one word to another in patterns –#3 (syntactic function of each word), is basic to the othersPaper 5: A Framework for Syntactic Translation. V. Yngve•Syntactic Translation–The central role of syntax1. Field of discourse•must use the relation of words in syntactic patterns as the key for finding which words refer to which field2. Recognition of coherent word groups•Idioms, noun compounds, and so on, are merely special patterns of words that stand out from more regular patterns•Modern Note:•Idioms are usually constituents and obey the rules of syntax•E.g. V O (VP) much more common than S V•Very few exceptions to the rules of syntax: by and large, by the byPaper 5: A Framework for Syntactic Translation. V. Yngve•Syntactic Translation–The central role of syntax4. Selectional relations between open class items–Selectional relations between words that are syntactic related–Modern Note:•E.g. subject of wake up must be animate5. Antecedents–The relationship of a word to its antecedent is essentially a syntactic relationship–Modern Note:•Binding Theory? C-command relation–Mary saw herself–*Mary’s father saw herselfPaper 5: A Framework for Syntactic Translation. V. Yngve•The Framework (Hypothetical)Paper 5: A Framework for Syntactic Translation. V. Yngve•Specifier = structural description = “message” or transition language•Tree-to-tree transfer model•Stored Knowledge–G1, E and G2 are declarative descriptions•Not procedural–Procedural information in R.R., T.R. and C.R. boxes•“Earlier estimates that the amount of storage necessary for syntactic information may be of the same order of magnitude as the amount of storage required for a dictionary have not been revised.”Paper 5: A Framework for Syntactic Translation. V. Yngve•Construction (C.R.)–Presumably does things like re-orders the constituents for English word order SVO (vs. German word order V2)Paper 5: A Framework for Syntactic Translation. V. Yngve•Specifiers–The specifier of a sentence represents that sentence as a series of choices within the limited range of choices prescribed by the grammar of the language•Sentence: affirmative or negative•Subject: modified by relative clause or not•Finite Verb: person, number, tense•Verb: which class–Ambiguity•Unspecified coordinatesPaper 5: A Framework for Syntactic Translation. V. Yngve•Specifiers–Narrow vs. broad specifiers–Examples:•Auxiliary verb can - present, past (narrow)• will be able to … future (broad)• has been able to … perfective (broad)•Auxiliary verb must - present (narrow)•had to - past•Present tense for future time•“He is coming soon”–Two type of specifiers for each language•5 step translation procedurePaper 5: A Framework for Syntactic Translation. V. Yngve•Recognition–The question of how we understand a sentence is a valid one for linguists, and it may have an answer different from the answer of how we produce a sentence. –But it appears that the description of a language is more easily couched in terms of synthesis of sentences than in terms of analysis of sentences. The reason is clear. •A description in terms of synthesis is straightforward and unambiguous. It is a one-to-one mapping of specifiers into sentences.•But a description in terms of analysis runs into all of the ambiguities of language that are caused by the chance overlapping of difference patterns: a sentence may be understandable in terms of two or more different specifiers.–Description in terms of analysis will probably not be available until after we have the more easily obtained descriptions in terms of synthesis.Paper 5: A Framework for Syntactic Translation. V. Yngve•Transfer of Structure–The real compromises in translation reside in these center boxes. It is here that the difficult and perhaps often impossible matching of sentences in different languages is undertaken.–But the problems associated with the center box are not peculiar to mechanical translation.–Human translators also face the very same problems.–The only difference is that at present the human translators are able