Statistical Machine Translation Models Research Articles

Efficient incremental training using a novel NMT-SMT hybrid framework for translation of low-resource languages.

The data-hungry statistical machine translation (SMT) and neural machine translation (NMT) models offer state-of-the-art results for languages with abundant data resources. However, extensive research is imperative to make these models perform equally well for low-resource languages. This paper proposes a novel approach to integrate the best features of the NMT and SMT systems for improved translation performance of low-resource English-Tamil language pair. The suboptimal NMT model trained with the small parallel corpus translates the monolingual corpus and selects only the best translations, to retrain itself in the next iteration. The proposed method employs the SMT phrase-pair table to determine the best translations, based on the maximum match between the words of the phrase-pair dictionary and each of the individual translations. This repeating cycle of translation and retraining generates a large quasi-parallel corpus, thus making the NMT model more powerful. SMT-integrated incremental training demonstrates a substantial difference in translation performance as compared to the existing approaches for incremental training. The model is strengthened further by adopting a beam search decoding strategy to produce k best possible translations for each input sentence. Empirical findings prove that the proposed model with BLEU scores of 19.56 and 23.49 outperforms the baseline NMT with scores 11.06 and 17.06 for Eng-to-Tam and Tam-to-Eng translations, respectively. METEOR score evaluation further corroborates these results, proving the supremacy of the proposed model.

Statistical machine translation for Indic languages

Abstract Statistical Machine Translation (SMT) systems use various probabilistic and statistical Natural Language Processing (NLP) methods to automatically translate from one language to another language while retaining the originality of the context. This paper aims to discuss the development of bilingual SMT models for translating English into fifteen low-resource Indic languages (ILs) and vice versa. The process to build the SMT model is described and explained using a workflow diagram. Samanantar and OPUS corpus are utilized for training, and Flores200 corpus is used for fine-tuning and testing purposes. The paper also highlights various preprocessing methods used to deal with corpus noise. The Moses open-source SMT toolkit is being investigated for the system’s development. The impact of distance-based reordering and Morpho-syntactic Descriptor Bidirectional Finite-State Encoder (msd-bidirectional-fe) reordering on ILs is compared in the paper. This paper provides a comparison of SMT models with Neural Machine Translation (NMT) for ILs. All the experiments assess the translation quality using standard metrics such as BiLingual Evaluation Understudy, Rank-based Intuitive Bilingual Evaluation Score, Translation Edit Rate, and Metric for Evaluation of Translation with Explicit Ordering. From the result, it is observed that msd-bidirectional-fe reordering performs better than the distance-based reordering model for ILs. It is also noticed that even though the IL-English and English-IL systems are trained using the same corpus, the former performs better for all the evaluation metrics. The comparison between SMT and NMT shows that across various languages, SMT performs better in some cases, while NMT outperforms in others.

Automatic translation from English to Amazigh using transformer learning

Due to the lack of parallel data, to our knowledge, no study has been conducted on the Amazigh-English language pair, despite the numerous machine translation studies completed between major European language pairs. We decided to utilize the neural machine translation (NMT) method on a parallel corpus of 137,322 sentences. The attention-based encoder-decoder architecture is used to construct statistical machine translation (SMT) models based on Moses, as well as NMT models using long short-term memory (LSTM), gated recurrent units (GRU), and transformers. Various outcomes were obtained for each strategy after several simulations: 80.7% accuracy was achieved using the statistical approach, 85.2% with the GRU model, 87.9% with the LSTM model, and 91.37% with the transformer.

On the use of statistical machine translation for suggesting variable names for decompiled code: The Pharo case

Adequately selecting variable names is a difficult activity for practitioners. In 2018, Jaffe et al. proposed the use of statistical machine translation (SMT) to suggest descriptive variable names for decompiled code. A large corpus of decompiled C code was used to train the SMT model. Our paper presents the results of a partial replication of Jaffe’s experiment. We apply the same technique and methodology to a dataset made of code written in the Pharo programming language. We selected Pharo since its syntax is simple – it fits on half of a postcard – and because the optimizations performed by the compiler are limited to method scope. Our results indicate that SMT may recover between 8.9% and 69.88% of the variable names depending on the training set. Our replication concludes that: (i) the accuracy depends on the code similarity between the training and testing sets; (ii) the simplicity of the Pharo syntax and the satisfactory decompiled code alignment have a positive impact on predicting variable names; and (iii) a relatively small code corpus is sufficient to train the SMT model, which shows the applicability of the approach to less popular programming languages. Additionally, to assess SMT’s potential in improving original variable names, ten Pharo developers reviewed 400 SMT name suggestions, with four reviews per variable. Only 15 suggestions (3.75%) were unanimously viewed as improvements, while 45 (11.25%) were perceived as improvements by at least two reviewers, highlighting SMT’s limitations in providing suitable alternatives.

Malayalam Natural Language Processing: Challenges in Building a Phrase-Based Statistical Machine Translation System

Statistical Machine Translation (SMT) is a preferred Machine Translation approach to convert the text in a specific language into another by automatically learning translations using a parallel corpus. SMT has been successful in producing quality translations in many foreign languages, but there are only a few works attempted in South Indian languages. The article discusses on experiments conducted with SMT for Malayalam language and analyzes how the methods defined for SMT in foreign languages affect a Dravidian language, Malayalam. The baseline SMT model does not work for Malayalam due to its unique characteristics like agglutinative nature and morphological richness. Hence, the challenge is to identify where precisely the SMT model has to be modified such that it adapts the challenges of the language peculiarity into the baseline model and give better translations for English to Malayalam translation. The alignments between English and Malayalam sentence pairs, subjected to the training process in SMT, plays a crucial role in producing quality output translation. Therefore, this work focuses on improving the translation model of SMT by refining the alignments between English–Malayalam sentence pairs. The phrase alignment algorithms align the verb and noun phrases in the sentence pairs and develop a new set of alignments for the English–Malayalam sentence pairs. These alignment sets refine the alignments formed from Giza++ produced as a result of EM training algorithm. The improved Phrase-Based SMT model trained using these refined alignments resulted in better translation quality, as indicated by the AER and BLUE scores.

Reordering of Source Side for a Factored English to Manipuri SMT System

Similar languages with massive parallel corpora are readily implemented by large-scale systems using either Statistical Machine Translation (SMT) or Neural Machine Translation (NMT). Translations involving low-resource language pairs with linguistic divergence have always been a challenge. We consider one such pair, English-Manipuri, which shows linguistic divergence and belongs to the low resource category. For such language pairs, SMT gets better acclamation than NMT. However, SMT’s more prominent phrase- based model uses groupings of surface word forms treated as phrases for translation. Therefore, without any linguistic knowledge, it fails to learn a proper mapping between the source and target language symbols. Our model adopts a factored model of SMT (FSMT3*) with a part-of-speech (POS) tag as a factor to incorporate linguistic information about the languages followed by hand-coded reordering. The reordering of source sentences makes them similar to the target language allowing better mapping between source and target symbols. The reordering also converts long-distance reordering problems to monotone reordering that SMT models can better handle, thereby reducing the load during decoding time. Additionally, we discover that adding a POS feature data enhances the system’s precision. Experimental results using automatic evaluation metrics show that our model improved over phrase-based and other factored models using the lexicalised Moses reordering options. Our FSMT3* model shows an increase in the automatic scores of translation result over the factored model with lexicalised phrase reordering (FSMT2) by an amount of 11.05% (Bilingual Evaluation Understudy), 5.46% (F1), 9.35% (Precision), and 2.56% (Recall), respectively.

Improved Unsupervised Statistical Machine Translation via Unsupervised Word Sense Disambiguation for a Low-Resource and Indic Languages

ABSTRACT Besides word order, word choice is a key stumbling block for machine translation (MT) in morphologically rich languages due to homonyms and polysemous difficulties. On the other hand, un-translated/improperly translated words are a severe issue for Statistical Machine Translation (SMT) models. The quantity of parallel training corpus has limited unsupervised SMT (USMT) systems. Still, current research lines have successfully trained SMT systems in an unsupervised manner using monolingual data alone. However, there is still a need to enhance the translation quality of the MT output due to unaligned and improperly sensed words. This problem is addressed by incorporating unsupervised Word Sense Disambiguation (WSD) into the decoding phase of USMT. The work provided a compendium of SMT systems for five translation tasks, i.e. En→Indic languages for the WMT test dataset and evaluated on BLEU and METEOR evaluation metrics. The studies were performed on En→Hi, En→Kn, En→Ta, En→Te, and En→Be tasks and showed an improvement in BLEU points by 2.3, 2.68, 0.78, 2.32, and 1.79, respectively, and METEOR points by 1.07, 1.34, 0.72, 0.693, and 1.191, respectively, over the baseline model.

Is the Corpus Ready for Machine Translation? A Case Study with Python to Pseudo-Code Corpus.

The availability of data is the driving force behind most of the state-of-the-art techniques for machine translation tasks. Understandably, this availability of data motivates researchers to propose new techniques and claim about the superiority of their techniques over the existing ones by using suitable evaluation measures. However, the performance of underlying learning algorithms can be greatly influenced by the correctness and the consistency of the corpus. We present our investigations for the relevance of a publicly available python to pseudo-code parallel corpus for automated documentation task, and the studies performed using this corpus. We found that the corpus had many visible issues like overlapping of instances, inconsistency in translation styles, incompleteness, and misspelled words. We show that these discrepancies can significantly influence the performance of the learning algorithms to the extent that they could have caused previous studies to draw incorrect conclusions. We performed our experimental study using statistical machine translation and neural machine translation models. We have recorded a significant difference (sim 10% on BLEU score) in the models’ performance after removing the issues from the corpus.

Improving Neural Machine Translation by Transferring Knowledge from Syntactic Constituent Alignment Learning

Statistical machine translation (SMT) models rely on word-, phrase-, and syntax-level alignments. But neural machine translation (NMT) models rarely explicitly learn the phrase- and syntax-level alignments. In this article, we propose to improve NMT by explicitly learning the bilingual syntactic constituent alignments. Specifically, we first utilize syntactic parsers to induce syntactic structures of sentences, and then we propose two ways to utilize the syntactic constituents in a perceptual (not adversarial) generator-discriminator training framework. One way is to use them to measure the alignment score of sentence-level training examples, and the other is to directly score the alignments of constituent-level examples generated with an algorithm based on word-level alignments from SMT. In our generator-discriminator framework, the discriminator is pre-trained to learn constituent alignments and distinguish the ground-truth translation from the fake ones, while the generative translation model is fine-tuned to receive the alignment knowledge and to generate translations that best approximate the true ones. Experiments and analysis show that the learned constituent alignments can help improve the translation results.

Knowledge Distillation: A Method for Making Neural Machine Translation More Efficient

Neural machine translation (NMT) systems have greatly improved the quality available from machine translation (MT) compared to statistical machine translation (SMT) systems. However, these state-of-the-art NMT models need much more computing power and data than SMT models, a requirement that is unsustainable in the long run and of very limited benefit in low-resource scenarios. To some extent, model compression—more specifically state-of-the-art knowledge distillation techniques—can remedy this. In this work, we investigate knowledge distillation on a simulated low-resource German-to-English translation task. We show that sequence-level knowledge distillation can be used to train small student models on knowledge distilled from large teacher models. Part of this work examines the influence of hyperparameter tuning on model performance when lowering the number of Transformer heads or limiting the vocabulary size. Interestingly, the accuracy of these student models is higher than that of the teachers in some cases even though the student model training times are shorter in some cases. In a novel contribution, we demonstrate for a specific MT service provider that in the post-deployment phase, distilled student models can reduce emissions, as well as cost purely in monetary terms, by almost 50%.

Low Resource Neural Machine Translation: Assamese to/from Other Indo-Aryan (Indic) Languages

Machine translation (MT) systems have been built using numerous different techniques for bridging the language barriers. These techniques are broadly categorized into approaches like Statistical Machine Translation (SMT) and Neural Machine Translation (NMT). End-to-end NMT systems significantly outperform SMT in translation quality on many language pairs, especially those with the adequate parallel corpus. We report comparative experiments on baseline MT systems for Assamese to other Indo-Aryan languages (in both translation directions) using the traditional Phrase-Based SMT as well as some more successful NMT architectures, namely basic sequence-to-sequence model with attention, Transformer, and finetuned Transformer. The results are evaluated using the most prominent and popular standard automatic metric BLEU (BiLingual Evaluation Understudy), as well as other well-known metrics for exploring the performance of different baseline MT systems, since this is the first such work involving Assamese. The evaluation scores are compared for SMT and NMT models for the effectiveness of bi-directional language pairs involving Assamese and other Indo-Aryan languages (Bangla, Gujarati, Hindi, Marathi, Odia, Sinhalese, and Urdu). The highest BLEU scores obtained are for Assamese to Sinhalese for SMT (35.63) and the Assamese to Bangla for NMT systems (seq2seq is 50.92, Transformer is 50.01, and finetuned Transformer is 50.19). We also try to relate the results with the language characteristics, distances, family trees, domains, data sizes, and sentence lengths. We find that the effect of the domain is the most important factor affecting the results for the given data domains and sizes. We compare our results with the only existing MT system for Assamese (Bing Translator) and also with pairs involving Hindi.

Generating Chinese Classical Poems with Statistical Machine Translation Models

This paper describes a statistical approach to generation of Chinese classical poetry and proposes a novel method to automatically evaluate poems. The system accepts a set of keywords representing the writing intents from a writer and generates sentences one by one to form a completed poem. A statistical machine translation (SMT) system is applied to generate new sentences, given the sentences generated previously. For each line of sentence a specific model specially trained for that line is used, as opposed to using a single model for all sentences. To enhance the coherence of sentences on every line, a coherence model using mutual information is applied to select candidates with better consistency with previous sentences. In addition, we demonstrate the effectiveness of the BLEU metric for evaluation with a novel method of generating diverse references.

Hybrid System Combination Framework for Uyghur–Chinese Machine Translation

Both the statistical machine translation (SMT) model and neural machine translation (NMT) model are the representative models in Uyghur–Chinese machine translation tasks with their own merits. Thus, it will be a promising direction to combine the advantages of them to further improve the translation performance. In this paper, we present a hybrid framework of developing a system combination for a Uyghur–Chinese machine translation task that works in three layers to achieve better translation results. In the first layer, we construct various machine translation systems including SMT and NMT. In the second layer, the outputs of multiple systems are combined to leverage the advantage of SMT and NMT models by using a multi-source-based system combination approach and the voting-based system combination approaches. Moreover, instead of selecting an individual system’s combined outputs as the final results, we transmit the outputs of the first layer and the second layer into the final layer to make a better prediction. Experiment results on the Uyghur–Chinese translation task show that the proposed framework can significantly outperform the baseline systems in terms of both the accuracy and fluency, which achieves a better performance by 1.75 BLEU points compared with the best individual system and by 0.66 BLEU points compared with the conventional system combination methods, respectively.

Improving Transformer‐Based Neural Machine Translation with Prior Alignments

Transformer is a neural machine translation model which revolutionizes machine translation. Compared with traditional statistical machine translation models and other neural machine translation models, the recently proposed transformer model radically and fundamentally changes machine translation with its self‐attention and cross‐attention mechanisms. These mechanisms effectively model token alignments between source and target sentences. It has been reported that the transformer model provides accurate posterior alignments. In this work, we empirically prove the reverse effect, showing that prior alignments help transformer models produce better translations. Experiment results on Vietnamese‐English news translation task show not only the positive effect of manually annotated alignments on transformer models but also the surprising outperformance of statistically constructed alignments reinforced with the flexibility of token‐type selection over manual alignments in improving transformer models. Statistically constructed word‐to‐lemma alignments are used to train a word‐to‐word transformer model. The novel hybrid transformer model improves the baseline transformer model and transformer model trained with manual alignments by 2.53 and 0.79 BLEU, respectively. In addition to BLEU score, we make limited human judgment on translation results. Strong correlation between human and machine judgment confirms our findings.

Transliterating Nôm Scripts into Vietnamese National Scripts using Statistical Machine Translation

Nôm scripts were used as the Vietnamese writing system from the 10th century to the early 20th century. During this period, Nôm scripts were the means to record a broad range of historical events, literary works, medical knowledge, as well as wisdom of many other domains. Unfortunately, since hardly any native Vietnamese speaker can read Nôm scripts nowadays, these valuable documents have not been fully harnessed. To address this gap, it is necessary to build an automatic transliteration system that can support us in decoding the ancient scripts and gaining knowledge of our Vietnamese ancestors. This study focuses on categorizing and reviewing the current progress on the Statistical Machine Translation (SMT) approaches to transliterate Nôm scripts into Vietnamese national scripts. In this paper, we discuss the differences between Nôm scripts and Vietnamese national scripts, systematically compare SMT models in transliterating Nôm scripts into Vietnamese national scripts, as well as having a thorough outlook on several promising research directions.

Analyzing Subword Techniques to Improve English to Sinhala Neural Machine Translation

Neural machine translation (NMT) is a remarkable approach which performs much better than the Statistical machine translation (SMT) models when there is an abundance of parallel corpus. However, vanilla NMT is primarily based upon word-level with a fixed vocabulary. Therefore, low resource morphologically rich languages such as Sinhala are mostly affected by the out of vocabulary (OOV) and Rare word problems. Recent advancements in subword techniques have opened up opportunities for low resource communities by enabling open vocabulary translation. In this paper, we extend our recently published state-of-the-art EN-SI translation system using the transformer and explore standard subword techniques on top of it to identify which subword approach has a greater effect on English Sinhala language pair. Our models demonstrate that subword segmentation strategies along with the state-of-the-art NMT can perform remarkably when translating English sentences into a rich morphology language regardless of a large parallel corpus.

Adaptation in Statistical Machine Translation for Low-resource Domains in English-Vietnamese Language

Adaptation in Statistical Machine Translation for Low-resource Domains in English-Vietnamese Language

Improving neural machine translation through phrase-based soft forced decoding

Compared to traditional statistical machine translation (SMT), such as phrase-based machine translation (PBMT), neural machine translation (NMT) often sacrifices adequacy for the sake of fluency. We propose a method to combine the advantages of traditional SMT and NMT by exploiting an existing phrase-based SMT model to compute the phrase-based decoding cost for an NMT output and then using this cost to rerank the n-best NMT outputs. The main challenge in implementing this approach is that NMT outputs may not be in the search space of the standard phrase-based decoding algorithm, because the search space of PBMT is limited by the phrase-based translation rule table. We propose a phrase-based soft forced decoding algorithm, which can always successfully find a decoding path for any NMT output. We show that using the phrase-based decoding cost to rerank the NMT outputs can successfully improve translation quality on four different language pairs.

Revisiting Back-Translation for Low-Resource Machine Translation Between Chinese and Vietnamese

Back-translation (BT) has been widely used and become one of standard techniques for data augmentation in Neural Machine Translation (NMT), BT has proven to be beneficial for improving the performance of translation effectively, especially for low-resource scenarios. While most previous works related to BT mainly focus on European languages with high relatedness, few of them study less-related languages in other areas around the world. In this paper, we choose the language pair with less relatedness in Asia: Chinese and Vietnamese, to investigate the impacts of BT on extremely low-resource machine translation between them. We first discuss the similarities and differences between the two languages, then evaluate and compare the effects of different sizes of back-translated data on NMT and Statistical Machine Translation (SMT) models for Chinese-Vietnamese and Vietnamese-Chinese, with both character-based and word-based settings, and conduct further analysis on the translation outputs from several aspects. Some conclusions from previous works are partially confirmed and we also draw some new findings and conclusions, which are beneficial to understand BT further and deeper for translation between less-related low-resource languages.

Research on statistical machine translation model based on deep neural network

With the increase of translation demand, the advancement of information technology, the development of linguistic theories and the progress of natural language understanding models in artificial intelligence research, machine translation has gradually gained worldwide attention. However, at present, machine translation research still has problems such as insufficient bilingual data and lack of effective feature representation, which affects the further improvement of key modules of machine translation such as word alignment, sequence adjustment and translation modelling. The effect of machine translation is still unsatisfactory. As a new machine learning method, deep neural network can automatically learn abstract feature representation and establish a complex mapping relationship between input and output signals, which provides a new idea for statistical machine translation research. Firstly, the multi-layer neural network and the undirected probability graph model are combined, and the similarity and context information of vocabulary are effectively utilized to model the word alignment more fully, and the word alignment model named NNWAM is constructed. Secondly, the low dimension will be used. The feature representation is combined with other features into a linearly ordered pre-ordering model to construct the pre-ordering model named NNPR. Finally, the word alignment model and the pre-ordering model are combined in the same deep neural network framework to form DNNAPM, a statistical machine translation model based on deep neural networks. The experimental results show that the statistical machine translation model based on deep neural network has better effect, faster convergence and better reliability than the comparison model algorithm.