Eisele, A. and Dorre, J. (1988). Unification of Disjunctive Feature Descriptions. In Proceedings of the 26th Annual Meeting of the ACL (ACL-88), Bu#alo, NY.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....[47] which resulted from an effort to bring together the work of Ait Kaci, Kasper Rounds, and Johnson. The constraint solving algorithm given here requires transformation into disjunctive normal form, which will usually cause an exponential blow up in size. Kasper [21, 20] and Dorre and Eisele [11] have proposed better unification algorithms for feature terms that try to avoid pushing up unions as much as possible. A new constraint solving method of Dorre and Eisele [9] introduces so called distributed disjunctions and works on feature constraints rather than feature terms. Kasper [22] ....

A. Eisele and J. Dorre. Unification of disjunctive feature descriptions. In Proceedings pages 286--294, Buffalo, New York, 1988.

....for Maher complete axiomatization of the algebra of rational trees. From a practical point of view the fact that the satisfiability problem is NPhard tends to manifest itself in a dramatic way in practical applications motivated several specialized algorithms to minimize this problem [Kas87,ED88,MK91] In [DV92] it was argued that any practical approach to the satisfiability problem should use factorization techniques to reduce the size of the input formulae to which any complete algorithm for satisfiability is applied, since such factorization can reduce by an exponential factor the ....

A. Eisele and J. Dorre. Unification of disjunctive feature descriptions. In 26th Annual Meeting of the Association for Computational Linguistics, Bu#alo, New York, 1988.

....van Noord [20] compares left corner and chart parsers that do apply ambiguity packing with a GLR parser without ambiguity packing. In Proceedings of IWPT 2000, Trento, Italy, February 2000, Pages 147 158 f structures has received quite a bit of attention in the literature over the last decade [4] [11] 12] 13] 6] While the two issues are related, we focus here on how to achieve optimal packing of c structures, so that unification operations do not have to be re executed due to the later detection of an additional local ambiguity. While there is much to be gained from improved packing ....

A. Eisele and J. Dorre. Unification of Disjunctive Feature Descriptions. In Proceedings of the 26th Annual Meeting of the ACL (ACL-88), Buffalo, NY, 1988.

....(P 4 OE 4 ) P 3 P 4 OE 6 ) 30) This approach would find smaller nogoods earlier, which would allow combinations of constraints that depended on those nogoods to be ignored, since the contexts would already be known to be inconsistent. 6. 4 Eisele and Dorre s techniques Eisele and Dorre[2] developed an algorithm for taking Karttunen s notion of disjunctive values a little further. Their algorithm allows disjunctive values to be unified with reentrant structures. The algorithm correctly detects such cases and lifts the disjunction due to reentrancy . They give the following ....

Eisele, A. and Dorre, J. (1988). Unification of Disjunctive Feature Descriptions. Proceedings of the 26th Annual Meeting of the ACL. Buffalo, New York.

....results to the logic of feature structures. In establishing these results, it is not our claim that Horn clause constraints are all that is necessary. There is already a substantial body of research on unification in the presence of disjunctive constraints on feature structures, of which [Kas87] [ED88], DE90] and [Str91] are but a sample. But all of these works investigate disjunction within the context of ordinary feature structures. Our position is that Horn clause constrained feature structures, rather than ordinary feature structures, ought to be considered as a natural starting point for ....

....not a theoretical limitation at all) and viewing such clauses as rules with disjunctive consequents, as illustrated by the form identified in 2.12. oe 1 oe 2 : oe k 1 ) ae 1 ae 2 : ae k 2 ) rf) Particularly, it is our conjecture that many of the techniques identified in [Kas87] [ED88], DE90] and [Str91] may be directly extended to the clausal framework. A great potential advantage of such an approach is that Horn constraints would be handled by our Algorithm 3, while only the disjunctive components would be handled specially. This would guarantee a tractable unification on ....

Eisele, A. and Dorre, J., "Unification of disjunctive feature descriptions," in: Proceedings of the 26th Annual Meeting of the ACL, pp. 286--294, 1988.

....Linguistics community over the past few years. From a practical point of view, the fact that the satisfiability problem (in these domains) is NP hard tends to manifest itself in a dramatic way in practical applications, motivated several specialized algorithms to minimize this problem [Kas87, ED88, MK91] In [DV92] it was argued that any practical approach to the satisfiability problem should use factorization techniques to reduce the size of the input formulae to which any complete algorithm for satisfiability is applied, since such factorization can reduce by an exponential factor the ....

A. Eisele and J. Dorre. Unification of disjunctive feature descriptions. In 26th Annual Meeting of the Association for Computational Linguistics, Buffalo, New York, 1988.

....describe exactly the same sets as variable free feature terms, thus bridging the gap between Johnson s and Rounds and Kasper s formalisms. However, feature terms with variables allow for exponentially more succinct descriptions, a fact that is important for efficient unification algorithms [12, 6]. We will also introduce quantification for feature terms with variables and present a quantifier elimination algorithm that computes for every feature term an equivalent quantifier free feature term, thus showing that even with quantification no more sets can be described and consistency and ....

....with variables and quantification can be translated into an equivalent feature term without variables. Thus variables 25 and quantification do not enable us to describe more sets, but just provide for more succinct descriptions, which is of crucial importance for efficient unification algorithms [12, 6]. We will show that the consistency problem for feature terms with variables but without quantifications is still in NP. However, if quantifications are present, we do not know whether the consistency problem remains in NP. To accommodate variables and quantifications, we extend the abstract ....

[Article contains additional citation context not shown here]

A. Eisele and J. Dorre. Unification of disjunctive feature descriptions. In Proceedings of the 26th Annual Meeting of the ACL, State University of New York at Buffalo, pages 286--294, Buffalo, New York, 1988.

.... the number of minimal satisfiers for a formula is exponential in the size of the formula in the worst case and satisfiability of a formula is NP complete (Kasper and Rounds 1986) Relatively efficient practical algorithms for dealing with disjunctions have been developed by Kasper (1987) and Eisele and Dorre (1988). Another option that is being explored is the utilization of total typing as discussed above, for managing memory allocation and improving the speed of both unification and the unwinding of information upon backtracking. The features values themselves could then be retrieved automatically without ....

Eisele, A. and Dorre, J. (1988). Unification of disjunctive feature descriptions. In Proceedings of the 26th Annual Conference of the Association for Computational Linguistics, Buffalo, New York.

.... u top(l 1 ) f ; Gammag) 1) The straightforward result of the unification top(l 1 ) X : f ; Gammag l 2 ) X : f ; Gammag) is incorrect, since it contains the term top(l 1 ) l 2 ) Gamma) as one possible extension, which is contradictory to the first unificand (see also Eisele and Dorre [10]) One possible solution is to expand the disjunction to the greatest common prefix of all addresses of a global coreference (cf. 10] Because we want terms to be able to share the 2 A similar system is introduced in Smolka [15] and Nebel and Smolka [14] same subterm this method cannot be ....

.... since it contains the term top(l 1 ) l 2 ) Gamma) as one possible extension, which is contradictory to the first unificand (see also Eisele and Dorre [10] One possible solution is to expand the disjunction to the greatest common prefix of all addresses of a global coreference (cf. [10]) Because we want terms to be able to share the 2 A similar system is introduced in Smolka [15] and Nebel and Smolka [14] same subterm this method cannot be employed. Instead we decided to use distributed disjunctions. The fundamental idea behind distributed disjunctions is that the ....

Andreas Eisele and Jochen Dorre. Unification of disjunctive feature descriptions. In 26 th Annual Meeting of the Association for Computational Linguistics, pages 186--194, Buffalo (NY), 1988.

....a significant amount of time is spent performing unification, optimizing feature structures for unification should increase the performance of these systems. Many algorithms for efficient unification of feature structures with dependent disjunctions have been proposed (Maxwell and Kaplan, 1989; Eisele and Dorre, 1990; Gerdemann, 1991; Stromback, 1992; Griffith, 1996) However, all of these algorithms suffer from a common problem: their performance is highly determined by their inputs. All of these algorithms will perform at their best when their dependent disjunctions interact as little as possible, but if ....

Eisele, A. and J. Dorre. 1990. Unification of Disjunctive Feature Descriptions IBM Wissenschaftliches Zentrum, Institute fur Wissensbasierte Systeme tech. report 124.

.... ) X) u top(l 1 ) f ; Gammag) 1) The straightforward result of the unification top(l 1 ) X : f ; Gammag l 2 ) X : f ; Gammag) is incorrect, since it contains the term top(l 1 ) l 2 ) Gamma) as one possible extension, which is contradictory to the first unificand (see also Eisele Dorre [8]) One possible solution is to expand the disjunction to the greatest common prefix of all addresses of an global coreference (cf. 8] Because we want terms to be able to share the same subterm this method cannot be employed. Instead we decided to use distributed disjunctions. The ....

.... is incorrect, since it contains the term top(l 1 ) l 2 ) Gamma) as one possible extension, which is contradictory to the first unificand (see also Eisele Dorre [8] One possible solution is to expand the disjunction to the greatest common prefix of all addresses of an global coreference (cf. [8]) Because we want terms to be able to share the same subterm this method cannot be employed. Instead we decided to use distributed disjunctions. The fundamental idea behind distributed disjunctions is that the problem of global coreferences can be solved by naming disjunctions. So the result ....

Andreas Eisele and Jochen Dorre. Unification of disjunctive feature descriptions. In 26 th Annual Meeting of the Association for Computational Linguistics, pages 186--194, Buffalo (NY), 1988.

....unification (Tomabeci, 1992) Optimizing feature structures for unification therefore potentially increases the performance of these systems. Many algorithms for efficient unification of feature structures with both normal and dependent disjunctions have been proposed, cf. Kasper, 1987b; Eisele and Dorre, 1988; V eronis, 1992) for algorithms designed for normal disjunctions and (Maxwell III and Kaplan, 1989; Eisele and Dorre, 1990; Backofen et al. 1991; Gerdemann, 1991; Stromback, 1992) for algorithms designed for dependent disjunctions) However, all of these algorithms suffer from a common problem: ....

Eisele, A. and Dorre, J. (1988). Unification of disjunctive feature descriptions.

No context found.

Eisele, A. and Dorre, J. (1988). Unification of Disjunctive Feature Descriptions. In Proceedings of the 26th Annual Meeting of the ACL (ACL-88), Bu#alo, NY.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC