Combining Morpheme-based Machine Translation with Post-processing Morpheme Prediction
Clifton, Ann and Sarkar, Anoop

Article Structure

Abstract

This paper extends the training and tuning regime for phrase-based statistical machine translation to obtain fluent translations into morphologically complex languages (we build an English to Finnish translation system).

Translation and Morphology

Languages with rich morphological systems present significant hurdles for statistical machine translation (SMT), most notably data sparsity, source-target asymmetry, and problems with automatic evaluation.

Models 2.1 Baseline Models

We set up three baseline models for comparison in this work.

Experimental Results

For all of the models built in this paper, we used the Europarl version 3 corpus (Koehn, 2005) English-Finnish training data set, as well as the standard development and test data sets.

Related Work

The work on morphology in MT can be grouped into three categories, factored models, segmented translation, and morphology generation.

Conclusion and Future Work

We found that using a segmented translation model based on unsupervised morphology induction and a model that combined morpheme segments in the translation model with a postprocessing morphology prediction model gave us better BLEU scores than a word-based baseline.

Topics

translation model

Appears in 21 sentences as: Translation Model (1) translation model (19) translation models (2)
In Combining Morpheme-based Machine Translation with Post-processing Morpheme Prediction
  1. Our main contributions are: 1) the introduction of the notion of segmented translation where we explicitly allow phrase pairs that can end with a dangling morpheme, which can connect with other morphemes as part of the translation process, and 2) the use of a fully segmented translation model in combination with a postprocessing morpheme prediction system, using unsupervised morphology induction.
    Page 1, “Translation and Morphology”
  2. Morphology can express both content and function categories, and our experiments show that it is important to use morphology both within the translation model (for morphology with content) and outside it (for morphology contributing to fluency).
    Page 1, “Translation and Morphology”
  3. Our second baseline is a factored translation model (Koehn and Hoang, 2007) (called Factored), which used as factors the word, “stem”1 and suffix.
    Page 2, “Models 2.1 Baseline Models”
  4. performance of unsupervised segmentation for translation, our third baseline is a segmented translation model based on a supervised segmentation model (called Sup), using the hand-built Omorfi morphological analyzer (Pirinen and Lis-tenmaa, 2007), which provided slightly higher BLEU scores than the word-based baseline.
    Page 2, “Models 2.1 Baseline Models”
  5. For segmented translation models , it cannot be taken for granted that greater linguistic accuracy in segmentation yields improved translation (Chang et al., 2008).
    Page 2, “Models 2.1 Baseline Models”
  6. Rather, the goal in segmentation for translation is instead to maXimize the amount of leXical content-carrying morphology, while generalizing over the information not helpful for improving the translation model .
    Page 2, “Models 2.1 Baseline Models”
  7. However, translation models based upon either Paramor alone or the combined systems output could not match the word-based baseline, so we concentrated on Morfessor.
    Page 2, “Models 2.1 Baseline Models”
  8. Figure 1(a) gives the full model overview for all the variants of the segmented translation model (supervised/ unsupervised; with and without the Unsup L—match procedure).
    Page 3, “Models 2.1 Baseline Models”
  9. Morphology generation as a postprocessing step allows major vocabulary reduction in the translation model , and allows the use of morphologically targeted features for modeling inflection.
    Page 3, “Models 2.1 Baseline Models”
  10. sider the morphology in translation since it is removed prior to training the translation model .
    Page 3, “Models 2.1 Baseline Models”
  11. (a) Segmented Translation Model
    Page 4, “Models 2.1 Baseline Models”

See all papers in Proc. ACL 2011 that mention translation model.

See all papers in Proc. ACL that mention translation model.

Back to top.

BLEU

Appears in 17 sentences as: BLEU (18)
In Combining Morpheme-based Machine Translation with Post-processing Morpheme Prediction
  1. Automatic evaluation measures for MT, BLEU (Papineni et al., 2002), WER (Word Error Rate) and PER (Position Independent Word Error Rate) use the word as the basic unit rather than morphemes.
    Page 1, “Translation and Morphology”
  2. Our proposed approaches are significantly better than the state of the art, achieving the highest reported BLEU scores on the English-Finnish Europarl version 3 dataset.
    Page 2, “Translation and Morphology”
  3. performance of unsupervised segmentation for translation, our third baseline is a segmented translation model based on a supervised segmentation model (called Sup), using the hand-built Omorfi morphological analyzer (Pirinen and Lis-tenmaa, 2007), which provided slightly higher BLEU scores than the word-based baseline.
    Page 2, “Models 2.1 Baseline Models”
  4. All the BLEU scores reported are for lowercase evaluation.
    Page 5, “Experimental Results”
  5. m-BLEU 1dicates that the segmented output was evaluated gainst a segmented version of the reference (this Leasure does not have the same correlation with hu-Lan judgement as BLEU ).
    Page 5, “Experimental Results”
  6. No Uni indicates the seg-Lented BLEU score without unigrams.
    Page 5, “Experimental Results”
  7. .on of m-BLEU score (Luong et al., 2010) where 1e BLEU score is computed by comparing the 3gmented output with a segmented reference ranslation.
    Page 5, “Experimental Results”
  8. 1troduced in this paper using standard word-ased lowercase BLEU , WER and PER.
    Page 5, “Experimental Results”
  9. Model BLEU WER TER
    Page 6, “Experimental Results”
  10. Table 3: Test Scores: lowercase BLEU , WER and TER.
    Page 6, “Experimental Results”
  11. The >|< indicates a statistically significant improvement of BLEU score over the Baseline model.
    Page 6, “Experimental Results”

See all papers in Proc. ACL 2011 that mention BLEU.

See all papers in Proc. ACL that mention BLEU.

Back to top.

CRF

Appears in 12 sentences as: CRF (12)
In Combining Morpheme-based Machine Translation with Post-processing Morpheme Prediction
  1. A conditional random field ( CRF ) (Lafferty et al., 2001) defines the conditional probability as a linear score for each candidate y and a global normalization term:
    Page 3, “Models 2.1 Baseline Models”
  2. However, the output 31* from the CRF decoder is still only a sequence of abstract suffix tags.
    Page 3, “Models 2.1 Baseline Models”
  3. The abstract suffix tags are extracted from the unsupervised morpheme learning process, and are carefully designed to enable CRF training and decoding.
    Page 4, “Models 2.1 Baseline Models”
  4. The output from the MT system is then used as input to the CRF model.
    Page 4, “Models 2.1 Baseline Models”
  5. The CRF model was trained on a ~210,000 Finnish sentences, consisting of ~1.5 million tokens; the 2,000 sentence Europarl test set consisted of 41,434 stem tokens.
    Page 4, “Models 2.1 Baseline Models”
  6. 4Note that unlike Section 2.2 we do not use Unsup L—match because when evaluating the CRF model on the suffix prediction task it obtained 95.61% without using Unsup L—match and 82.99% when using Unsup L—match.
    Page 4, “Models 2.1 Baseline Models”
  7. Post-Process 2: CRF _ Lagguagfe Model _ M r h | ' sur ace orm mapping 0 p 0 Ogy Generatlon stem+morph1+ +morph2
    Page 4, “Models 2.1 Baseline Models”
  8. This resulted in 44 possible label outputs per stem which was a reasonable sized tag-set for CRF training.
    Page 4, “Models 2.1 Baseline Models”
  9. The CRF was trained on monolingual features of the segmented text for suffix prediction, where t is the current token:
    Page 4, “Models 2.1 Baseline Models”
  10. After CRF based recovery of the suffix tag sequence, we use a bigram language model trained on a full segmented version on the training data to recover the original vowels.
    Page 4, “Models 2.1 Baseline Models”
  11. original training data: koskevaa mietintoa kasitellaan segmentation: koske+ +va+ +a mietinto+ +5 kasi+ +te+ +115+ +a+ +n (train bigram language model with mapping A = { a, 'a map final suflia‘ to abstract tag-set: koske+ +va+ +A mietinto+ +A kasi+ +te+ +11a+ +a+ +n (train CRF model to predict the final suffix) peeling of final suflia‘: koske+ +va+ mietinto+ kasi+ +te+ +11a+ +a+ (train SMT model on this transformation of training data) (a) Training
    Page 5, “Models 2.1 Baseline Models”

See all papers in Proc. ACL 2011 that mention CRF.

See all papers in Proc. ACL that mention CRF.

Back to top.

segmentation model

Appears in 10 sentences as: segmentation model (4) segmentation models (1) Segmented Model (1) segmented model (3) segmented model’s (1)
In Combining Morpheme-based Machine Translation with Post-processing Morpheme Prediction
  1. These are derived from the same unsupervised segmentation model used in other experiments.
    Page 2, “Models 2.1 Baseline Models”
  2. performance of unsupervised segmentation for translation, our third baseline is a segmented translation model based on a supervised segmentation model (called Sup), using the hand-built Omorfi morphological analyzer (Pirinen and Lis-tenmaa, 2007), which provided slightly higher BLEU scores than the word-based baseline.
    Page 2, “Models 2.1 Baseline Models”
  3. We therefore trained several different segmentation models , considering factors of granularity, coverage, and source-target symmetry.
    Page 2, “Models 2.1 Baseline Models”
  4. Morfessor uses minimum description length criteria to train a HMM-based segmentation model .
    Page 2, “Models 2.1 Baseline Models”
  5. Morphology generation models can use a variety of bilingual and contextual information to capture dependencies between morphemes, often more long-distance than what is possible using n-gram language models over morphemes in the segmented model .
    Page 3, “Models 2.1 Baseline Models”
  6. ‘able 2: Segmented Model Scores.
    Page 5, “Experimental Results”
  7. We find that when using a ood segmentation model , segmentation of the lorphologically complex target language im-roves model performance over an unsegmented aseline (the confidence scores come from boot-3rap resampling).
    Page 5, “Experimental Results”
  8. So, we ran the word-based baseline system, the segmented model (Unsup L—match), and the prediction model (CRF—LM) outputs, along with the reference translation through the supervised morphological analyzer Omorfi (Piri—nen and Listenmaa, 2007).
    Page 6, “Experimental Results”
  9. We saw that in both these categories, the CRF-LM model outperforms for precision, while the segmented model gets the best recall.
    Page 6, “Experimental Results”
  10. The goal of this experiment was to control the segmented model’s tendency to overfit by rewarding it for using correct whole-word forms.
    Page 8, “Related Work”

See all papers in Proc. ACL 2011 that mention segmentation model.

See all papers in Proc. ACL that mention segmentation model.

Back to top.

language model

Appears in 9 sentences as: language model (7) language models (3)
In Combining Morpheme-based Machine Translation with Post-processing Morpheme Prediction
  1. Morphology generation models can use a variety of bilingual and contextual information to capture dependencies between morphemes, often more long-distance than what is possible using n-gram language models over morphemes in the segmented model.
    Page 3, “Models 2.1 Baseline Models”
  2. is to take the abstract suffix tag sequence 31* and then map it into fully inflected word forms, and rank those outputs using a morphemic language model .
    Page 4, “Models 2.1 Baseline Models”
  3. After CRF based recovery of the suffix tag sequence, we use a bigram language model trained on a full segmented version on the training data to recover the original vowels.
    Page 4, “Models 2.1 Baseline Models”
  4. original training data: koskevaa mietintoa kasitellaan segmentation: koske+ +va+ +a mietinto+ +5 kasi+ +te+ +115+ +a+ +n (train bigram language model with mapping A = { a, 'a map final suflia‘ to abstract tag-set: koske+ +va+ +A mietinto+ +A kasi+ +te+ +11a+ +a+ +n (train CRF model to predict the final suffix) peeling of final suflia‘: koske+ +va+ mietinto+ kasi+ +te+ +11a+ +a+ (train SMT model on this transformation of training data) (a) Training
    Page 5, “Models 2.1 Baseline Models”
  5. We trained all of the Moses systems herein using the standard features: language model , reordering model, translation model, and word penalty; in addition to these, the factored experiments called for additional translation and generation features for the added factors as noted above.
    Page 5, “Experimental Results”
  6. For the language models, we used SRILM 5-gram language models (Stol-cke, 2002) for all factors.
    Page 5, “Experimental Results”
  7. koske+ +va+ +A mietinto+ +A kasi+ +te+ +11a+ +a+ +n language model disambiguation:
    Page 5, “Experimental Results”
  8. They use a segmented phrase table and language model along With the word-based versions in the decoder and in tuning a Finnish target.
    Page 8, “Related Work”
  9. In their work a segmented language model can score a translation, but cannot insert morphology that does not show source-side reflexes.
    Page 8, “Related Work”

See all papers in Proc. ACL 2011 that mention language model.

See all papers in Proc. ACL that mention language model.

Back to top.

BLEU score

Appears in 8 sentences as: BLEU score (4) BLEU scores (4)
In Combining Morpheme-based Machine Translation with Post-processing Morpheme Prediction
  1. Our proposed approaches are significantly better than the state of the art, achieving the highest reported BLEU scores on the English-Finnish Europarl version 3 dataset.
    Page 2, “Translation and Morphology”
  2. performance of unsupervised segmentation for translation, our third baseline is a segmented translation model based on a supervised segmentation model (called Sup), using the hand-built Omorfi morphological analyzer (Pirinen and Lis-tenmaa, 2007), which provided slightly higher BLEU scores than the word-based baseline.
    Page 2, “Models 2.1 Baseline Models”
  3. All the BLEU scores reported are for lowercase evaluation.
    Page 5, “Experimental Results”
  4. No Uni indicates the seg-Lented BLEU score without unigrams.
    Page 5, “Experimental Results”
  5. .on of m-BLEU score (Luong et al., 2010) where 1e BLEU score is computed by comparing the 3gmented output with a segmented reference ranslation.
    Page 5, “Experimental Results”
  6. The >|< indicates a statistically significant improvement of BLEU score over the Baseline model.
    Page 6, “Experimental Results”
  7. LM model is producing output translations that are more morphologically fluent than the word-based baseline and the segmented translation Unsup L—match system, even though the word choices lead to a lower BLEU score overall when compared to Unsup L—match.
    Page 7, “Experimental Results”
  8. We found that using a segmented translation model based on unsupervised morphology induction and a model that combined morpheme segments in the translation model with a postprocessing morphology prediction model gave us better BLEU scores than a word-based baseline.
    Page 9, “Conclusion and Future Work”

See all papers in Proc. ACL 2011 that mention BLEU score.

See all papers in Proc. ACL that mention BLEU score.

Back to top.

morphological analyzer

Appears in 7 sentences as: morphological analysis (3) morphological analyzer (4)
In Combining Morpheme-based Machine Translation with Post-processing Morpheme Prediction
  1. In fact, in our experiments, unsupervised morphology always outperforms the use of a hand-built morphological analyzer .
    Page 1, “Translation and Morphology”
  2. performance of unsupervised segmentation for translation, our third baseline is a segmented translation model based on a supervised segmentation model (called Sup), using the hand-built Omorfi morphological analyzer (Pirinen and Lis-tenmaa, 2007), which provided slightly higher BLEU scores than the word-based baseline.
    Page 2, “Models 2.1 Baseline Models”
  3. So, we ran the word-based baseline system, the segmented model (Unsup L—match), and the prediction model (CRF—LM) outputs, along with the reference translation through the supervised morphological analyzer Omorfi (Piri—nen and Listenmaa, 2007).
    Page 6, “Experimental Results”
  4. Segmented translation performs morphological analysis on the morphologically complex text for use in the translation model (Brown et al., 1993; Goldwater and McClosky, 2005; de Gispert and Marifio, 2008).
    Page 7, “Related Work”
  5. Previous work in segmented translation has often used linguistically motivated morphological analysis selectively applied based on a language-specific heuristic.
    Page 9, “Related Work”
  6. In order to help with replication of the results in this paper, we have run the various morphological analysis steps and created the necessary training, tuning and test data files needed in order to train, tune and test any phrase-based machine translation system with our data.
    Page 9, “Conclusion and Future Work”
  7. We would particularly like to thank the developers of the open-source Moses machine translation toolkit and the Omorfi morphological analyzer for Finnish which we used for our experiments.
    Page 9, “Conclusion and Future Work”

See all papers in Proc. ACL 2011 that mention morphological analyzer.

See all papers in Proc. ACL that mention morphological analyzer.

Back to top.

phrase-based

Appears in 6 sentences as: phrase-based (6)
In Combining Morpheme-based Machine Translation with Post-processing Morpheme Prediction
  1. This paper extends the training and tuning regime for phrase-based statistical machine translation to obtain fluent translations into morphologically complex languages (we build an English to Finnish translation system).
    Page 1, “Abstract”
  2. In this work, we propose to address the problem of morphological complexity in an English-to-Finnish MT task within a phrase-based translation framework.
    Page 1, “Translation and Morphology”
  3. We then trained the Moses phrase-based system (Koehn et al., 2007) on the segmented and marked text.
    Page 3, “Models 2.1 Baseline Models”
  4. In all the experiments conducted in this paper, we used the Moses5 phrase-based translation system (Koehn et al., 2007), 2008 version.
    Page 5, “Experimental Results”
  5. We also demonstrate that for Finnish (and possibly other agglutinative languages), phrase-based MT benefits from allowing the translation model access to morphological segmentation yielding productive morphological phrases.
    Page 9, “Conclusion and Future Work”
  6. In order to help with replication of the results in this paper, we have run the various morphological analysis steps and created the necessary training, tuning and test data files needed in order to train, tune and test any phrase-based machine translation system with our data.
    Page 9, “Conclusion and Future Work”

See all papers in Proc. ACL 2011 that mention phrase-based.

See all papers in Proc. ACL that mention phrase-based.

Back to top.

MT system

Appears in 5 sentences as: MT System (2) MT system (3)
In Combining Morpheme-based Machine Translation with Post-processing Morpheme Prediction
  1. Table 1 shows how morphemes are being used in the MT system .
    Page 3, “Models 2.1 Baseline Models”
  2. The first phase of our morphology prediction model is to train a MT system that produces morphologically simplified word forms in the target language.
    Page 3, “Models 2.1 Baseline Models”
  3. MT System Alignment:
    Page 4, “Models 2.1 Baseline Models”
  4. The output from the MT system is then used as input to the CRF model.
    Page 4, “Models 2.1 Baseline Models”
  5. stem+ +morph1+ V Y MT System stem+ +morph1t// Alignment: word word word
    Page 4, “Models 2.1 Baseline Models”

See all papers in Proc. ACL 2011 that mention MT system.

See all papers in Proc. ACL that mention MT system.

Back to top.

machine translation

Appears in 4 sentences as: machine translation (4)
In Combining Morpheme-based Machine Translation with Post-processing Morpheme Prediction
  1. This paper extends the training and tuning regime for phrase-based statistical machine translation to obtain fluent translations into morphologically complex languages (we build an English to Finnish translation system).
    Page 1, “Abstract”
  2. Languages with rich morphological systems present significant hurdles for statistical machine translation (SMT), most notably data sparsity, source-target asymmetry, and problems with automatic evaluation.
    Page 1, “Translation and Morphology”
  3. In order to help with replication of the results in this paper, we have run the various morphological analysis steps and created the necessary training, tuning and test data files needed in order to train, tune and test any phrase-based machine translation system with our data.
    Page 9, “Conclusion and Future Work”
  4. We would particularly like to thank the developers of the open-source Moses machine translation toolkit and the Omorfi morphological analyzer for Finnish which we used for our experiments.
    Page 9, “Conclusion and Future Work”

See all papers in Proc. ACL 2011 that mention machine translation.

See all papers in Proc. ACL that mention machine translation.

Back to top.

translation system

Appears in 4 sentences as: translation system (3) translation systems (1)
In Combining Morpheme-based Machine Translation with Post-processing Morpheme Prediction
  1. This paper extends the training and tuning regime for phrase-based statistical machine translation to obtain fluent translations into morphologically complex languages (we build an English to Finnish translation system ).
    Page 1, “Abstract”
  2. In all the experiments conducted in this paper, we used the Moses5 phrase-based translation system (Koehn et al., 2007), 2008 version.
    Page 5, “Experimental Results”
  3. For evaluation against segmented translation systems in segmented forms before word reconstruction, we also segmented the baseline system’s word-based output.
    Page 5, “Experimental Results”
  4. In order to help with replication of the results in this paper, we have run the various morphological analysis steps and created the necessary training, tuning and test data files needed in order to train, tune and test any phrase-based machine translation system with our data.
    Page 9, “Conclusion and Future Work”

See all papers in Proc. ACL 2011 that mention translation system.

See all papers in Proc. ACL that mention translation system.

Back to top.

segmentations

Appears in 4 sentences as: segmentations (4)
In Combining Morpheme-based Machine Translation with Post-processing Morpheme Prediction
  1. When tested against a human-annotated gold standard of linguistic morpheme segmentations for Finnish, this algorithm outperforms competing unsupervised methods, achieving an F—score of 67.0% on a 3 million sentence corpus (Creutz and Lagus, 2006).
    Page 2, “Models 2.1 Baseline Models”
  2. In order to get robust, common segmentations , we trained the segmenter on the 5000 most frequent words2; we then used this to segment the entire data set.
    Page 2, “Models 2.1 Baseline Models”
  3. Of the phrases that included segmentations (‘Morph’ in Table 1), roughly a third were ‘productive’, i.e.
    Page 3, “Models 2.1 Baseline Models”
  4. However, in phrases used while decoding the development and test data, roughly a quarter of the phrases that generated the translated output included segmentations , but of these, only a small fraction (6%) had a hanging morpheme; and while there are many possible reasons to account for this we were unable to find a single convincing cause.
    Page 3, “Models 2.1 Baseline Models”

See all papers in Proc. ACL 2011 that mention segmentations.

See all papers in Proc. ACL that mention segmentations.

Back to top.

translation task

Appears in 3 sentences as: translation task (3)
In Combining Morpheme-based Machine Translation with Post-processing Morpheme Prediction
  1. We show, using both automatic evaluation scores and linguistically motivated analyses of the output, that our methods outperform previously proposed ones and provide the best known results on the English-Finnish Europarl translation task .
    Page 1, “Abstract”
  2. Both of these approaches beat the state of the art on the English-Finnish translation task .
    Page 1, “Translation and Morphology”
  3. Using our proposed approach we obtain better scores than the state of the art on the English-Finnish translation task (Luong et al., 2010): from 14.82% BLEU to 15.09%, while using a
    Page 9, “Conclusion and Future Work”

See all papers in Proc. ACL 2011 that mention translation task.

See all papers in Proc. ACL that mention translation task.

Back to top.

Baseline system

Appears in 3 sentences as: Baseline system (1) baseline system (1) baseline system’s (1)
In Combining Morpheme-based Machine Translation with Post-processing Morpheme Prediction
  1. For our word-based Baseline system , we trained a word-based model using the same Moses system with identical settings.
    Page 5, “Experimental Results”
  2. For evaluation against segmented translation systems in segmented forms before word reconstruction, we also segmented the baseline system’s word-based output.
    Page 5, “Experimental Results”
  3. So, we ran the word-based baseline system , the segmented model (Unsup L—match), and the prediction model (CRF—LM) outputs, along with the reference translation through the supervised morphological analyzer Omorfi (Piri—nen and Listenmaa, 2007).
    Page 6, “Experimental Results”

See all papers in Proc. ACL 2011 that mention Baseline system.

See all papers in Proc. ACL that mention Baseline system.

Back to top.

state of the art

Appears in 3 sentences as: state of the art (3)
In Combining Morpheme-based Machine Translation with Post-processing Morpheme Prediction
  1. Both of these approaches beat the state of the art on the English-Finnish translation task.
    Page 1, “Translation and Morphology”
  2. Our proposed approaches are significantly better than the state of the art , achieving the highest reported BLEU scores on the English-Finnish Europarl version 3 dataset.
    Page 2, “Translation and Morphology”
  3. Using our proposed approach we obtain better scores than the state of the art on the English-Finnish translation task (Luong et al., 2010): from 14.82% BLEU to 15.09%, while using a
    Page 9, “Conclusion and Future Work”

See all papers in Proc. ACL 2011 that mention state of the art.

See all papers in Proc. ACL that mention state of the art.

Back to top.

phrase table

Appears in 3 sentences as: phrase table (3)
In Combining Morpheme-based Machine Translation with Post-processing Morpheme Prediction
  1. Table 1: Morpheme occurences in the phrase table and in translation.
    Page 3, “Models 2.1 Baseline Models”
  2. They use a segmented phrase table and language model along With the word-based versions in the decoder and in tuning a Finnish target.
    Page 8, “Related Work”
  3. Habash (2007) provides various methods to incorporate morphological variants of words in the phrase table in order to help recognize out of vocabulary words in the source language.
    Page 9, “Related Work”

See all papers in Proc. ACL 2011 that mention phrase table.

See all papers in Proc. ACL that mention phrase table.

Back to top.

bigram

Appears in 3 sentences as: bigram (2) bigrams (1)
In Combining Morpheme-based Machine Translation with Post-processing Morpheme Prediction
  1. After CRF based recovery of the suffix tag sequence, we use a bigram language model trained on a full segmented version on the training data to recover the original vowels.
    Page 4, “Models 2.1 Baseline Models”
  2. We used bigrams only, because the suffix vowel harmony alternation depends only upon the preceding phonemes in the word from which it was segmented.
    Page 4, “Models 2.1 Baseline Models”
  3. original training data: koskevaa mietintoa kasitellaan segmentation: koske+ +va+ +a mietinto+ +5 kasi+ +te+ +115+ +a+ +n (train bigram language model with mapping A = { a, 'a map final suflia‘ to abstract tag-set: koske+ +va+ +A mietinto+ +A kasi+ +te+ +11a+ +a+ +n (train CRF model to predict the final suffix) peeling of final suflia‘: koske+ +va+ mietinto+ kasi+ +te+ +11a+ +a+ (train SMT model on this transformation of training data) (a) Training
    Page 5, “Models 2.1 Baseline Models”

See all papers in Proc. ACL 2011 that mention bigram.

See all papers in Proc. ACL that mention bigram.

Back to top.

PDF available from www.aclweb.org.
Send us an email