Pairwise Neural Machine Translation Evaluation

Ivan Zapreev
Mind Map by Ivan Zapreev, updated more than 1 year ago
Ivan Zapreev
Created by Ivan Zapreev about 6 years ago
19
0

Description

Paper from Francisco Guzman, Shafiq Joty, Lluis Marquez and Preslav Nakov

Resource summary

Pairwise Neural Machine Translation Evaluation
  1. Introduction
    1. Automated Machine Translation (MT)
      1. Evaluation
        1. Needed
          1. Developing a new MT
            1. Comparing two MT
            2. Reference based MT
              1. Comparing the system output to one or more human reference tranlations
                1. Most Common
                  1. Compute
                    1. Absolute quality score
                      1. Computing similarity between the machine and human translation
                      2. Simplest case
                        1. Computing word N-gram matches between the translation and the reference
                          1. BLUE
                        2. More advanced
                          1. Take into account various aspects of linguistic similarity
                            1. Better correlation with human judjment
                      3. Human ranking
                        1. Can be used to train automatic metrics
                          1. Can be oriented to predict absolute scores
                            1. Using
                              1. Regression
                                1. ?
                                2. Ranking
                                  1. Special case
                                    1. Compare two hypotheses and referenec
                                      1. Decide which hypothesis is better
                                        1. Recent result
                                          1. Guzman
                                            1. Learning framework
                                              1. Using
                                                1. Preference kernel
                                                  1. ?
                                                  2. Vector machines (SVM)
                                                    1. Syntactic structures
                                                      1. Discourage-based structures
                                                      2. High computational costs
                                                        1. Training
                                                          1. Testing
                                                            1. Due to
                                                              1. Using convolution kernels
                                                                1. ?
                                                                2. Over complex structures
                                                                3. Simplification is needed!
                                    2. Research
                                      1. Framework for machine translation evaluation
                                        1. Novel!
                                          1. Goal
                                            1. Select a better translation from a pair of hypothesis, given the reference translation
                                            2. Using
                                              1. Neural networks
                                                1. Multi-layer
                                                  1. Input layer
                                                    1. Semantic info
                                                      1. Syntactic info
                                                        1. Lexical info
                                                        2. Hidden layer
                                                          1. Captures the interactions between the relevant input components
                                                        3. Models the interaction between
                                                          1. The two hypothesis
                                                            1. Reference and Hypothesis
                                                          2. Distributed vector representations
                                                            1. Used for storing
                                                              1. The Two hypothesis
                                                              2. Based on
                                                                1. Word embedding
                                                                  1. Sentence embedding
                                                                  2. Learned from
                                                                    1. Neural Networks
                                                                      1. Novel!
                                                                        1. Can be trained to optimize task-specific cost function
                                                                      2. Efficient
                                                                        1. Vector-based compression
                                                                          1. ?
                                                                    2. Experiments
                                                                      1. WMT12 metrics task
                                                                        1. Better results than by Guzman
                                                                          1. High correlation with human judjment
                                                                            1. Comparable with the best!
                                                                              1. DiscoTK
                                                                                1. Metric
                                                                                  1. Combination based
                                                                                    1. Much heavier
                                                                              2. Embeddings
                                                                                1. Syntactically oriented
                                                                                  1. Semantically oriented
                                                                                    1. Cumulative performance gains
                                                                                      1. Over
                                                                                        1. BLUE
                                                                                          1. NIST
                                                                                            1. METEOR
                                                                                              1. TER
                                                                                      2. Simplification is needed!
                                                                                      Show full summary Hide full summary

                                                                                      Similar

                                                                                      Evaluation of Explanations of Schizophrenia
                                                                                      Charlotte97
                                                                                      Nucleic Acids
                                                                                      Jessica Phillips
                                                                                      DNA Synthesis, Transcription, and Translation
                                                                                      badern
                                                                                      Evaluation: Social Causation Hypothesis as an Explanation for Schizophrenia
                                                                                      Katie Greensted
                                                                                      Evaluation of Conformity
                                                                                      littlestephie
                                                                                      Protein Synthesis: Transcription and Translation
                                                                                      itstinachen
                                                                                      Evaluation: Drug Treatment for Anorexia
                                                                                      Katie Greensted
                                                                                      Evaluation: Genetic Explanation of Anorexia Nervosa
                                                                                      Katie Greensted
                                                                                      Evaluation: Psychological Formulation
                                                                                      Katie Greensted
                                                                                      2.7 DNA Replication, Transcription and Translation
                                                                                      Steve Harton
                                                                                      surah Naba
                                                                                      hajraofficial14