II Regular Papers
Refine
Has Fulltext
- yes (15)
Year of publication
- 2008 (15)
Document Type
Language
- English (15)
Is part of the Bibliography
- no (15)
Institute
- Extern (15)
This article describes a HMM-based word-alignment method that can selectively enforce a contiguity constraint. This method has a direct application in the extraction of a bilingual terminological lexicon from a parallel corpus, but can also be used as a preliminary step for the extraction of phrase pairs in a Phrase-Based Statistical Machine Translation system. Contiguous source words composing terms are aligned to contiguous target language words. The HMM is transformed into a Weighted Finite State Transducer (WFST) and contiguity constraints are enforced by specific multi-tape WFSTs. The proposed method is especially suited when basic linguistic resources (morphological analyzer, part-of-speech taggers and term extractors) are available for the source language only.
We introduce and discuss a number of issues that arise in the process of building a finite-state morphological analyzer for Urdu, in particular issues with potential ambiguity and non-concatenative morphology. Our approach allows for an underlyingly similar treatment of both Urdu and Hindi via a cascade of finite-state transducers that transliterates the very different scripts into a common ASCII transcription system. As this transliteration system is based on the XFST tools that the Urdu/Hindi common morphological analyzer is also implemented in, no compatibility problems arise.
The emergence of information extraction (IE) oriented pattern engines has been observed during the last decade. Most of them exploit heavily finite-state devices. This paper introduces ExPRESS – a new extraction pattern engine, whose rules are regular expressions over flat feature structures. The underlying pattern language is a blend of two previously introduced IE oriented pattern formalisms, namely, JAPE, used in the widely known GATE system, and the unificationbased XTDL formalism used in SProUT. A brief and technical overview of ExPRESS, its pattern language and the pool of its native linguistic components is given. Furthermore, the implementation of the grammar interpreter is addressed too.
This paper describes a two-level formalism where feature structures are used in contextual rules. Whereas usual two-level grammars describe rational sets over symbol pairs, this new formalism uses tree structured regular expressions. They allow an explicit and precise definition of the scope of feature structures. A given surface form may be described using several feature structures. Feature unification is expressed in contextual rules using variables, like in a unification grammar. Grammars are compiled in finite state multi-tape transducers.
In this paper, we present a finite-state approach to constituency and therewith an analysis of coordination phenomena involving so-called non-constituents. We show that non-constituents can be seen as parts of fully-fledged constituents and therefore be coordinated in the same way. We have implemented an algorithm based on finite state automata that generates an LFG grammar assigning valid analyses to non-constituent coordination structures in the German language.
Finite state methods for natural language processing often require the construction and the intersection of several automata. In this paper, we investigate the question of determining the best order in which these intersections should be performed. We take as an example lexical disambiguation in polarity grammars. We show that there is no efficient way to minimize the state complexity of these intersections.
In this work an extension of CSSR algorithm using Maximum Entropy Models is introduced. Preliminary experiments to perform Named Entity Recognition with this new system are presented.
Morphological analyses based on word syntax approaches can encounter difficulties with long distance dependencies. The reason is that in some cases an affix has to have access to the inner structure of the form with which it combines. One solution is the percolation of features from ther inner morphemes to the outer morphemes with some process of feature unification. However, the obstacle of percolation constraints or stipulated features has lead some linguists to argue in favour of other frameworks such as, e.g., realizational morphology or parallel approaches like optimality theory. This paper proposes a linguistic analysis of two long distance dependencies in the morphology of Russian verbs, namely secondary imperfectivization and deverbal nominalization.We show how these processes can be reanalysed as local dependencies. Although finitestate frameworks are not bound by such linguistically motivated considerations, we present an implementation of our analysis as proposed in [1] that does not complicate the grammar or enlarge the network unproportionally.
We present an algorithm that computes a function that assigns consecutive integers to trees recognized by a deterministic, acyclic, finite-state, bottom-up tree automaton. Such function is called minimal perfect hashing. It can be used to identify trees recognized by the automaton. Its value may be seen as an index in some other data structures. We also present an algorithm for inverted hashing.
In the last years, statistical machine translation has already demonstrated its usefulness within a wide variety of translation applications. In this line, phrase-based alignment models have become the reference to follow in order to build competitive systems. Finite state models are always an interesting framework because there are well-known efficient algorithms for their representation and manipulation. This document is a contribution to the evolution of finite state models towards a phrase-based approach. The inference of stochastic transducers that are based on bilingual phrases is carefully analysed from a finite state point of view. Indeed, the algorithmic phenomena that have to be taken into account in order to deal with such phrase-based finite state models when in decoding time are also in-depth detailed.