Joint Generation of Transliterations from Multiple Representations

Yao, Lei; Kondrak, Grzegorz

doi:10.3115/v1/n15-1095

Cited by 5 publications

(6 citation statements)

References 14 publications

(14 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The data and the dictionaries can be accessed via https://www.hucompute.org/ressourcen/corpora so that our findings may be used as a starting point for related research. In future work, we intend to investigate more sophisticated combination techniques for combining outputs of several spell checkers, e.g., on the character-level, as done in Cortes et al (2014); Eger (2015d,c); Yao and Kondrak (2015). We also intend to evaluate neural-network based techniques in the present scenario (Sutskever et al, 2014;Yao and Zweig, 2015).…”

Section: Resultsmentioning

confidence: 99%

A Comparison of Four Character-Level String-to-String Translation Models for (OCR) Spelling Error Correction

Eger¹,

Brück²,

Mehler³

2016

The Prague Bulletin of Mathematical Linguistics

View full text Add to dashboard Cite

We consider the isolated spelling error correction problem as a specific subproblem of the more general string-to-string translation problem. In this context, we investigate four general string-to-string transformation models that have been suggested in recent years and apply them within the spelling error correction paradigm. In particular, we investigate how a simple ‘k-best decoding plus dictionary lookup’ strategy performs in this context and find that such an approach can significantly outdo baselines such as edit distance, weighted edit distance, and the noisy channel Brill and Moore model to spelling error correction. We also consider elementary combination techniques for our models such as language model weighted majority voting and center string combination. Finally, we consider real-world OCR post-correction for a dataset sampled from medieval Latin texts.

show abstract

Section: Resultsmentioning

confidence: 99%

A Comparison of Four Character-Level String-to-String Translation Models for (OCR) Spelling Error Correction

Eger¹,

Brück²,

Mehler³

2016

The Prague Bulletin of Mathematical Linguistics

View full text Add to dashboard Cite

show abstract

“…The JOINT system was trained on a held-out set composed of the outputs of the base systems generated for each source word. The second set of experiments followed the original design of Yao and Kondrak (2015), in which the supplemental data consists of transliterations of a source word in other languages. We extracted the supplemental transliterations from the NEWS 2015 Shared Task training and development sets for which English was the source language.…”

Section: Jointmentioning

confidence: 99%

“…Yao and Kondrak (2015) propose a JOINT generation approach that can incorporate multiple transliterations as input, and show that it outperforms the reranking approach of Bhargava and Kondrak (2012). The JOINT system is a modified version of DIRECTL+ that utilizes aligned supplemental transliterations to learn additional features.…”

Section: Jointmentioning

confidence: 99%

Multiple System Combination for Transliteration

Nicolai

Hauer

Salameh

et al. 2015

Proceedings of the Fifth Named Entity Workshop

Self Cite

View full text Add to dashboard Cite

We report the results of our experiments in the context of the NEWS 2015 Shared Task on Transliteration. We focus on methods of combining multiple base systems, and leveraging transliterations from multiple languages. We show error reductions over the best base system of up to 10% when using supplemental transliterations, and up to 20% when using system combination. We also discuss the quality of the shared task datasets.

show abstract

“…The work by Oh and Choi actually widened a new direction of research on transliteration started by Al-Onaizan and Knight [24] and later Bilac and Tanaka [81], of combining several sources and/or including supplemental transliterations. This has led to several recent works such as Kumaran et al [77], Bhargava and Kondrak (2011Kondrak ( , 2012 and Yao and Kondrak [82].…”

Section: Hybrid Approachesmentioning

confidence: 99%

Machine transliteration and transliterated text retrieval: a survey

Prabhakar

Pal

2018

Sādhanā

View full text Add to dashboard Cite

Users of the WWW across the globe are increasing rapidly. According to Internet live stats there are more than 3 billion Internet users worldwide today and the number of non-English native speakers is quite high there. A large proportion of these non-English speakers access the Internet in their native languages but use the Roman script to express themselves through various communication channels like messages and posts. With the advent of Web 2.0, user-generated content is increasing on the Web at a very rapid rate. A substantial proportion of this content is transliterated data. To leverage this huge information repository, there is a matching effort to process transliterated text. In this article, we survey the recent body of work in the field of transliteration. We start with a definition and discussion of the different types of transliteration followed by various deterministic and non-deterministic approaches used to tackle transliteration-related issues in machine translation and information retrieval. Finally, we study the performance of those techniques and present a comparative analysis of them.

show abstract

Joint Generation of Transliterations from Multiple Representations

Cited by 5 publications

References 14 publications

A Comparison of Four Character-Level String-to-String Translation Models for (OCR) Spelling Error Correction

A Comparison of Four Character-Level String-to-String Translation Models for (OCR) Spelling Error Correction

Multiple System Combination for Transliteration

Machine transliteration and transliterated text retrieval: a survey

Contact Info

Product

Resources

About