MT for Minority Languages Using Elicitation-Based Learning of Syntactic Transfer Rules

Probst, Katharina; Levin, Lori; Peterson, Erik; Lavie, Alon; Carbonell, Jaime G.

doi:10.1023/b:coat.0000021003.55041.fd

Cited by 28 publications

(28 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Unlike the other approaches used to automatically learn shallow-transfer rules from parallel corpora (Sánchez-Martínez and Forcada, 2009;Caseli et al, 2006;Probst et al, 2002), here all the bilingual phrase pairs are considered together when checking their reproducibility by the set of GATs obtained. We thus treat conflicting rules at a global level, while previous approaches treat them locally.…”

Section: Choosing the Most Appropriate Generalised Alignment Templatesmentioning

confidence: 99%

“…Fifth, the strategy applied in order to generalise bilingual phrase pairs to rules is clearly different. Their initial approach (Probst et al, 2002) consists of selecting the minimum set of rules which correctly translates the set of bilingual phrase pairs by following a greedy strategy based on merging pairs of rules, while ours selects the minimum set of rules by using a global strategy based on integer linear programming that is able to find the optimal solution. In her latest approach (Probst, 2005), a two-step procedure is followed for the generalisation problem.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

A generalised alignment template formalism and its application to the inference of shallow-transfer machine translation rules from scarce bilingual corpora

Sánchez-Cartagena¹,

Pérez-Ortiz

Sánchez-Martínez

2015

Computer Speech & Language

View full text Add to dashboard Cite

Statistical and rule-based methods are complementary approaches to machine translation (MT) that have different strengths and weaknesses. This complementarity has, over the last few years, resulted in the consolidation of a growing interest in hybrid systems that combine both data-driven and linguistic approaches. In this paper we address the situation in which the amount of bilingual resources that is available for a particular language pair is not sufficiently large to train a competitive statistical MT system, but the cost and slow development cycles of rule-based MT systems cannot be afforded either. In this context, we formalise a new method that uses scarce parallel corpora to automatically infer a set of shallowtransfer rules to be integrated into a rule-based MT system, thus avoiding the need for human experts to handcraft these rules.Our work is based on the alignment template approach to phrase-based statistical MT, but the definition of the alignment template is extended to encompass different generalisation levels. It is also greatly inspired by the work of Sánchez-Martínez and Forcada published in 2009 (Journal of Artificial Intelligence Research 34) in which alignment templates were also considered for shallow-transfer rule inference. However, our approach overcomes many relevant limitations of that work, principally those related to the inability to find the correct generalisation level for the alignment templates, and to select the subset of alignment templates that ensures an adequate segmentation of the input sentences by the rules eventually obtained. Unlike previous approaches in literature, our formalism does not require linguistic knowledge about the languages involved in the translation. Moreover, it is the first time that conflicts between rules are resolved by choosing the most appropriate ones according to a global minimisation function rather than proceeding in a pairwise greedy fashion.Experiments conducted using five different language pairs with the free/open-source rule-based MT platform Apertium show that translation quality significantly improves when compared to the method proposed by Sánchez-Martínez and Forcada (2009), and is close to that obtained using handcrafted rules. For some language pairs, our approach is even able to outperform them. Moreover, the resulting number of rules is considerably smaller, which eases human revision and maintenance.

show abstract

Section: Choosing the Most Appropriate Generalised Alignment Templatesmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

A generalised alignment template formalism and its application to the inference of shallow-transfer machine translation rules from scarce bilingual corpora

Sánchez-Cartagena¹,

Pérez-Ortiz

Sánchez-Martínez

2015

Computer Speech & Language

View full text Add to dashboard Cite

show abstract

“…As stated above, the user is currently responsible for the creation of rules, albeit with a natural, visual interface that often is able to set up the requisite input semantic structures automatically. As mentioned, we also seek to collaborate with researchers to enable semiautomatic generation of rules similar to what can be found in the Boas (McShane, et al, 2002), LinGO (Bender, et al, 2010), PAWS (Black and Black, 2009) and Avenue (Probst, et al, 2003) projects. Such extensions will make LA accessible to a larger pool of linguists and will shorten the time needed for documenting languages.…”

Section: Linguist's Assistantmentioning

confidence: 99%

Time to Change the "D" in "DEL"

Beale¹

2014

Proceedings of the 2014 Workshop on the Use of Computational Methods in the Study of Endangered Languages

View full text Add to dashboard Cite

The "D" in "DEL" stands for "documenting" -a code word for linguists that means the collection of linguistic data in audio and written form. The DEL (Documenting Endangered Languages) program run by the NSF and NEH is thus centered around building and archiving data resources for endangered languages. This paper is an argument for extending the 'D' to include "describing" languages in terms of lexical, semantic, morphological and grammatical knowledge. We present an overview of descriptive computational tools aimed at endangered languages along with a longer summary of two particular computer programs: Linguist's Assistant and Boas. These two programs, respectively, represent research in the areas of: A) computational systems capable of representing lexical, morphological and grammatical structures and using the resulting computational models for translation in a minority language context, and B) tools for efficiently and accurately acquiring linguistic knowledge. A hoped-for side effect of this paper is to promote cooperation between these areas of research in order to provide a total solution to describing endangered languages.

show abstract

“…A prime example of this latter point was the Avenue Project at Carnegie Mellon University (Probst, et al 2003). The Avenue project was a machine translation system oriented towards low-density languages.…”

Section: Research In Describing Endangered Languages: Knowledge Acquimentioning

confidence: 99%

“…As stated above, the user is currently responsible for the creation of rules, albeit with a natural, visual interface that often is able to set up the requisite input semantic structures automatically. As mentioned, we also seek to collaborate with researchers to enable semiautomatic generation of rules similar to what can be found in the Boas (McShane, et al, 2002), LinGO , PAWS (Black and Black, 2009) and Avenue (Probst, et al, 2003) projects. Such extensions will make LA accessible to a larger pool of linguists and will shorten the time needed for documenting languages.…”

Section: Linguist's Assistantmentioning

confidence: 99%

Proceedings of the 2014 Workshop on the Use of Computational Methods in the Study of Endangered Languages

2014

View full text Add to dashboard Cite

ii Preface Contemporary efforts to document the world's endangered languages-often going under the rubric of documentary linguistics-are dependent on the widespread availability of modern recording technologies, in particular digital audio and video recording devices and software to annotate the recordings that such devices produce. However, despite well over a decade of dedicated funding efforts aimed at the documentation of endangered languages, the technological landscape that supports the work of those involved in this research remains fragmented, and the promises of new technology remain largely unfulfilled. Moreover, the efforts of computer scientists, on the whole, are mostly disconnected from the day-to-day work of documentary linguists, making it difficult for the knowledge of each group to inform the other. On the one hand, this deprives documentary linguists of tools making use of the latest research results to speed up the time-consuming task of describing an underdocumented language. On the other hand, it severely limits the ability of computational linguists to test their methods on the full range of world's linguistic diversity.Despite the concerns listed above, recent efforts do indicate that there is significant potential in collaboration between computational linguists (and other computer scientists) and linguists working on endangered languages. For instance, machine labeling and active learning can make the process of textual analysis for low-resource languages more efficient, and state-of-the-art tools in grammar engineering can be applied at a relatively low cost to new languages that are typologically divergent from those that primarily informed their design. Moreover, new models of data collection based on the ubiquity of lowcost, networkable devices with recording capabilities, such as smartphones, show the extent to which the barriers to collecting significant amounts of primary data have fallen in recent years, and it has similarly been found that the pairing of crowdsourcing and machine translation techniques can yield useful results for low resource languages in a short time frame. Research along these latter lines, in particular, indicates that computationally-driven advances in the documentation of the world's languages may need to rely as much on clever engineering and user interface solutions as on methods for processing language data developed within computational linguistics proper, in a manner parallel to efforts in other domains that have considered how new online services can be used to facilitate computational linguistic research.A different set of activities within the documentary linguistics community involving the increasing use of open standards for encoding language data is also significant in this regard. For instance, in the last decade, standardized XML formats have become more widely used to encode text annotations and lexical data. This facilitates the reuse of documentary materials. Even in the absence of the use of such standards, significant results have been a...

show abstract

MT for Minority Languages Using Elicitation-Based Learning of Syntactic Transfer Rules

Cited by 28 publications

References 19 publications

A generalised alignment template formalism and its application to the inference of shallow-transfer machine translation rules from scarce bilingual corpora

A generalised alignment template formalism and its application to the inference of shallow-transfer machine translation rules from scarce bilingual corpora

Time to Change the "D" in "DEL"

Proceedings of the 2014 Workshop on the Use of Computational Methods in the Study of Endangered Languages

Contact Info

Product

Resources

About