Technical Reports

A List by Author: Luboš Popelínský

e-mail:
popel(a)fi.muni.cz
home page:
http://www.fi.muni.cz/usr/popelinsky/

Proceedings of the Third Learning Language in Logic Workshop

by Luboš Popelínský, Miloslav Nepil, September 2001, 66 pages.

FIMU-RS-2001-08. Available as Postscript, PDF.

Abstract:

The 3rd Learning Language in Logic (LLL) workshop held in August 8-9 in Strasbourg was the follow-up of the previous LLL workshops held in 1999 in Bled, Slovenia, and in 2000 in Lisboa, Portugal. The LLL workshop took place as a joint workshop of ILP`2001 conference. The program of the workshop split into two parts. Work-in-progress papers presented in Aug 8 afternoon are contained in this report.

On Disambiguation in Czech Corpora

by Luboš Popelínský, Tomáš Pavelek, Tomáš Ptáčník, October 2000, 26 pages.

FIMU-RS-2000-07. Available as Postscript, PDF.

Abstract:

Lemma disambiguation means finding the basic word form, typically nominative singular for nouns or infinitive for verbs. We developed a multistrategy method for lemma disambiguation of unannotated text. The method is based on a combination of inductive logic programming and instance-based learning. We present results of the most important subtasks of lemma disambiguation for Czech language. Although no expert knowledge on Czech grammar has been used the accuracy reaches 90% with a fraction of words remaining ambiguous. We also display first results of tag disambiguation.

WiM: A Study on the Top-Down ILP Program

by Luboš Popelínský, August 1995, 18 pages.

FIMU-RS-95-03. Available as Postscript, PDF.

Abstract:

In the area of the inductive synthesis of logic programs it is the small number of examples which is crucial. We show that the classical MIS-like architecture can be adapted using techniques described in ILP literature so that we reach very good results if to compare with other ILP systems. We describe the top-down ILP program WiM and the results obtained through it. WiM needs from 2 to 4 examples for most of the ILP benchmark predicates. Even though it is interactive, not more that one membership query is enough to receive the correct target program. WiM has higher efficiency of learning as well as smaller dependency on the quality of the example set in comparison to some of ILP programs. The quality of learning has been tested both on good examples and on randomly chosen example sets.

On Biases in Inductive Data Engineering

by Jana Kuklová, Luboš Popelínský, March 1995, 7 pages.

FIMU-RS-95-01. Available as Postscript, PDF.

Abstract:

An utilization of inductive reasoning [Flene94] in the database area looks promising [Agar93], [DeRae92], [DeRae93], [Flach93], [Kivi92], [Savnik93], [Manni93]. Inductive data engineering, as introduced in [Flach93], means a process of restructuring database by means of induction.

In our work we focus on exploitation of inductive logic programming for database schema design [Roll92]. We propose modifications of INDEX, described in [Flach93], namely new biases for narrowing search space, as well as stopping criterion

In this article the case of single relation is described.

Responsible contact: unix(atsign)fi(dot)muni(dot)cz