Phrase linguistic classification and generalization for improving statistical...Hiroshi Matsumoto
De Gispert, Adrià. "Phrase linguistic classification and generalization for improving statistical machine translation." Proceedings of the ACL Student Research Workshop. Association for Computational Linguistics, 2005.
1. The document discusses the relationship between Hilbert systems (H) and natural deduction systems (N), showing how H maps to N via lambda abstraction.
2. It introduces Martin-Löf type theory (ML-ITT) and explains how propositions as types allows representing proofs as terms. ML-ITT can interpret both H and N through this correspondence.
3. Several works are cited that explore how ML-ITT can be viewed as an interpretation of set theory through universes, and how the Curry-Howard correspondence and lambda calculus are fundamental to ML-ITT.
Neural Models for Information Retrieval
Bhaskar Mitra, Nick Craswell
(Submitted on 3 May 2017)
Neural ranking models for information retrieval (IR) use shallow or deep neural networks to rank search results in response to a query. Traditional learning to rank models employ machine learning techniques over hand-crafted IR features. By contrast, neural models learn representations of language from raw text that can bridge the gap between query and document vocabulary. Unlike classical IR models, these new machine learning based approaches are data-hungry, requiring large scale training data before they can be deployed. This tutorial introduces basic concepts and intuitions behind neural IR models, and places them in the context of traditional retrieval models. We begin by introducing fundamental concepts of IR and different neural and non-neural approaches to learning vector representations of text. We then review shallow neural IR methods that employ pre-trained neural term embeddings without learning the IR task end-to-end. We introduce deep neural networks next, discussing popular deep architectures. Finally, we review the current DNN models for information retrieval. We conclude with a discussion on potential future directions for neural IR.
Phrase linguistic classification and generalization for improving statistical...Hiroshi Matsumoto
De Gispert, Adrià. "Phrase linguistic classification and generalization for improving statistical machine translation." Proceedings of the ACL Student Research Workshop. Association for Computational Linguistics, 2005.
1. The document discusses the relationship between Hilbert systems (H) and natural deduction systems (N), showing how H maps to N via lambda abstraction.
2. It introduces Martin-Löf type theory (ML-ITT) and explains how propositions as types allows representing proofs as terms. ML-ITT can interpret both H and N through this correspondence.
3. Several works are cited that explore how ML-ITT can be viewed as an interpretation of set theory through universes, and how the Curry-Howard correspondence and lambda calculus are fundamental to ML-ITT.
Neural Models for Information Retrieval
Bhaskar Mitra, Nick Craswell
(Submitted on 3 May 2017)
Neural ranking models for information retrieval (IR) use shallow or deep neural networks to rank search results in response to a query. Traditional learning to rank models employ machine learning techniques over hand-crafted IR features. By contrast, neural models learn representations of language from raw text that can bridge the gap between query and document vocabulary. Unlike classical IR models, these new machine learning based approaches are data-hungry, requiring large scale training data before they can be deployed. This tutorial introduces basic concepts and intuitions behind neural IR models, and places them in the context of traditional retrieval models. We begin by introducing fundamental concepts of IR and different neural and non-neural approaches to learning vector representations of text. We then review shallow neural IR methods that employ pre-trained neural term embeddings without learning the IR task end-to-end. We introduce deep neural networks next, discussing popular deep architectures. Finally, we review the current DNN models for information retrieval. We conclude with a discussion on potential future directions for neural IR.
This document discusses a study that integrated multiple rule-based machine translation engines into a hybrid system using Moses. The system architecture combines the phrase tables from Moses and each RBMT system. The RBMT outputs are aligned and their phrase tables concatenated with the Moses phrase table. The tuning process adjusts weights for the additional columns from the RBMT phrase tables. Results showed BLEU score improvements from combining rule-based and data-driven approaches into a hybrid machine translation system.
This document summarizes a paper that explores relearning a rule-based machine translation (RBMT) system using statistical methods. It compares the performance of the original SYSTRAN RBMT system, a relearnt statistical model of SYSTRAN called SYSTRAN Relearnt, and a baseline statistical model called SYSTRAN Relearnt-0. The models are trained without parallel corpora by using SYSTRAN translations. Evaluation shows SYSTRAN Relearnt achieves 5 BLEU points higher than the baseline by using a real English language model and tuning set. Error analysis of 100 sentences identifies common error types between the systems like missing words, extra words, and translation choices to discriminate the nature and training of
This paper proposes a method for example-based machine translation that combines syntactic transfer with statistical models. The method uses transfer rules to construct the target language syntactic tree structure from the source language. It then uses a statistical generation module to select the best word sequence based on language and translation models. The method is evaluated on a travel domain corpus, with the combined approach outperforming a baseline of example-based transfer alone in terms of BLEU, NIST and human evaluation.
The document summarizes an English-Japanese example-based machine translation system developed by Microsoft Research (MSR-MT) that uses abstract linguistic representations. MSR-MT combines rule-based and statistical techniques with example-based transfer. It first parses sentence pairs into logical forms (LFs) and then extracts mappings between the LFs to create a bilingual knowledge base. New sentences are translated by matching their LFs to the knowledge base. An evaluation found MSR-MT performed comparably to a commercial system on a technical domain, suggesting example-based MT can achieve good results using semantic representations and alignment rules.
The document summarizes the BLEU method for automatically evaluating machine translation systems. BLEU calculates n-gram precision between a candidate translation and multiple reference translations, with modifications to address weaknesses. It combines the average logarithm of modified n-gram precisions with a brevity penalty for translations longer than references. Evaluation tests on multiple translation systems found BLEU scores reliably distinguished system quality and correlated well with human judgements.
This document describes a statistical approach to machine translation. It discusses using probability to determine the most likely source sentence S given a target sentence T. It presents methods for computing language model probabilities, translation probabilities, and searching for the optimal S. Two pilot experiments are described to estimate parameters for the translation model from bilingual text data. Evaluation of the second experiment showed the decoded sentences were either exact, alternate, different, wrong or ungrammatical compared to the reference translations.
Approach to japanese english automatic translation by Susumu KunoHiroshi Matsumoto
1. The document describes a machine translation system for translating Japanese text to another language.
2. It involves automatic input editing, segmentation, syntactical analysis, and output editing with transformation.
3. The system handles characteristics of Japanese text like having no spaces between words and using kanji characters, by segmenting the text into components and replacing kanji with word tokens.
4. Pre-edited(予備編集):
• input is marked to indicate prefixes, suffixes
• 入力値の接頭辞、接尾辞などに指標を付ける作業
Control(入力値コントロール):
• to control the vocabularies and grammatical structure
• 語順の入れ替え/差し替え作業
Sublanguage(部分言語):
• MT system specialized in a specific field
5. Bilingual systems(二か国語):
• Designed for two particular languages;特定の二言語用に設計されたもの
• Unidirectional(一方向性):
Source Language(対象言語):SL Target Language(目的言語):TL
• Bidirectional(両方向性):
SL TL
• Most are unidirectional
多くは一方向となる
Multilingual systems(多言語):
• Designed for more than a single pair of languages
• 多言語に対応するように設計されたもの
• Most are bidirectional
• 両方向が多い
10. In Analysis:解析
Morphology(形態論):
• Identification of word endings, word compounds
• -ing, -ed などの接尾語、airport, housetopなどの複合語の認証
Syntax(統語論):
• Identification of phrase structures, dependency subordination
• フレーズ、従属関係の認証
Semantics(意味論):
• Resolution of lexical and structural ambiguities
• 語彙、構造の曖昧解消
In Synthesis:統合(生成)
Semantics:
• Selection of appropriate compatible lexical and structural forms
• 語彙、構造の妥当選別
Syntax:
• Generation of required phrase and sentence structures
• フレーズ、構文の生成
Morphology:
• Generation of correct word forms
• 正当語形生成
12. The idea:
• to reuse examples of already existing translations as the basis for a new
translation.
• 「ある文と似た文は元の文と同じように訳される」という原理
• ある文を翻訳することをそれとよく似た文の翻訳を見つけ、それを模倣す
ることによって行う
Three stages (3段階プロセス):
• Matching: matching the input against database
• マッチング:入力とデータベースの照合
• Alignment: identifying corresponding translation fragments
• アライメント:対応した翻訳部位との判定
• Recombination: recombining these fragments
• リコンビネーション:つなぎ合わせ
13. Steps:
• First to align phrases, word groups and individual words of the
parallel texts
• フレーズ、語彙群、個々の語彙との位置合わせ
• Calculate the probabilities of correspondence of words in SL, TL
• SL,TLとの一致度合の確率の計算