539 research outputs found
Sort by
Approximating a Deep-Syntactic Metric for MT Evaluation and Tuning
No full textSemPOS is an automatic metric of machine
translation quality for Czech and English focused
on content words. It correlates well
with human judgments but it is computationally
costly and hard to adapt to other languages
because it relies on a deep-syntactic
analysis of the system output and the reference.
To remedy this, we attempt at approximating
SemPOS using only tagger output and
a few heuristics. At a little expense in correlation
to human judgments, we can evaluate
MT systems much faster. Additionally, we describe
our submission to the Tunable Metrics
Task in WMT11
TamilTB - Tamil Dependency Treebank
No full textTamilTB is a first published syntactically annotated corpus of Tamil. TamilTB will allow a more rapid development of language technologies for Tamil (especially parsing and morphological analysis)
Czech-Slovak Parallel Corpora
No full textThe paper describes suitable sources for creating Czech-Slovak
parallel corpora, including our procedure of creating plain text parallel
corpora from various data sources. We attempt to address the pros and
cons of various types of data sources, especially when they are used in
machine translation. Some results of machine translation from Czech to
Slovak based on the acquired corpora are also given
Forms Wanted: Training SMT on Monolingual Data
No full textWe propose and evaluate a simple technique of "reverse self-training" for
statistical machine translation. The technique allows to extend target-side
vocabulary of the MT system using target-side monolingual data and it is
especially aimed at translation to morphologically rich languages
Generating Czech Word Forms in MT: From System Combination to Black Art
No full textThe results of my experiments with MT system combination aimed at better coverage of target-side word forms
Influence of Parser Choice on Dependency-Based MT
No full textAccuracy of dependency parsers is one of the key factors limiting the quality of dependency-based machine translation. This paper deals with the influence of various dependency parsing approaches (and also different training data size) on the overall performance of an English-to-Czech dependency-based statistical translation system implemented in the Treex framework. We also study the relationship between parsing accuracy in terms of unlabeled attachment score and machine translation quality in terms of BLEU
A Grain of Salt for the WMT Manual Evaluation
Get PDFThe Workshop on Statistical Machine Translation (WMT) has become one of ACL's
flagship workshops, held annually since 2006. In addition to soliciting papers
from the research community, WMT also features a shared translation task for
evaluating MT systems. This shared task is notable for having manual evaluation
as its cornerstone.
The Workshop's overview paper, playing a descriptive and administrative role, reports
the main results of the evaluation without delving deep into analyzing those results.
The aim of this paper is to investigate and explain some interesting idiosyncrasies
in the reported results, which only become apparent when performing a more thorough
analysis of the collected annotations. Our analysis sheds some light on how the
reported results should (and should not) be interpreted, and also gives rise to some helpful
recommendation for the organizers of WMT
TamilTB - An Effort Towards Building a Treebank for Tamil
No full textThis talk is aimed at presenting our ongoing effort to build a PDT style dependency treebank for Tamil language. The talk will outline the annotation scheme and annotation at morphological and surface syntax layers. Various issues such as ambiguous structures, NP compounding, coordination phenomena and clitics with respect to the treebank annotation will be discussed. Our ultimate goal in this project is to develop a feature rich parsing framework for Tamil, thus we also present the results we obtained in automatic parsing (rule based & corpus based) using the developed resources. Some problematic issues in Tamil parsing will also be discussed
Treex – an open-source framework for natural language processing
No full textThe present paper describes Treex (formerly TectoMT), a multi-purpose open-source
framework for developing Natural Language Processing applications. It facilitates the development by exploiting a wide range of software modules already integrated in Treex, such as tools for sentence segmentation, tokenization, morphological analysis,
part-of-speech tagging, shallow and deep syntax parsing, named entity recognition, anaphora resolution, sentence synthesis, word-level alignment of parallel corpora, and other tasks.
The most elaborate application of Treex is
an English-Czech machine translation system with transfer on deep syntactic (tectogrammatical) layer. Besides research, Treex is used
for teaching purposes and helps students to implement morphological and syntactic analyzers of foreign languages in a very short time
Jak nachytat strojový překlad na švestkách
No full textAn introduction to the task of machine translation, current methods and their inherent limitations for those interested in computer science