Dependency and Phrasal Parsers of the Czech Language: A Comparison - - PowerPoint PPT Presentation

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions

Dependency and Phrasal Parsers of the Czech Language: A Comparison

Aleˇ s Hor´ ak1, Tom´ aˇ s Holan2, Vladim´ ır Kadlec1, Vojtˇ ech Kov´ aˇ r1

Faculty of Informatics, Masaryk University, Brno, Czech Republic {hales,xkadlec,xkovar3}@fi.muni.cz Faculty of Mathematics and Physics, Charles University, Prague, Czech Republic Tomas.Holan@mff.cuni.cz

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions

Outline

1 Motivation 2 Compared Parsers 3 Differences of the Parses 4 Results 5 Conclusions and Future Directions

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions

Motivation

Several parsing systems for Czech based on different theories No comparison between different techniques has been done so far

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions

Motivation

Several parsing systems for Czech based on different theories No comparison between different techniques has been done so far

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions

Compared Parsers

Dependency stochastic parsers – Prague parsers Phrasal rule based parser – synt

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions

Compared Parsers

Dependency stochastic parsers – Prague parsers Phrasal rule based parser – synt

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Prague Parsers

Prague Parsers

McD McDonnald’s maximum spanning tree parser, COL Collins’s parser adapted for PDT, ZZ ˇ Zabokrtsk´ y’s rule-based dependency parser, AN Holan’s parser ANALOG, L2R, R2L, L2R3, R2L3 Holan’s pushdown parsers, CP Holan’s and ˇ Zabokrtsk´ y ’s combining parser.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Prague Parsers

Prague Parsers

McD McDonnald’s maximum spanning tree parser, COL Collins’s parser adapted for PDT, ZZ ˇ Zabokrtsk´ y’s rule-based dependency parser, AN Holan’s parser ANALOG, L2R, R2L, L2R3, R2L3 Holan’s pushdown parsers, CP Holan’s and ˇ Zabokrtsk´ y ’s combining parser.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Prague Parsers

Prague Parsers

McD McDonnald’s maximum spanning tree parser, COL Collins’s parser adapted for PDT, ZZ ˇ Zabokrtsk´ y’s rule-based dependency parser, AN Holan’s parser ANALOG, L2R, R2L, L2R3, R2L3 Holan’s pushdown parsers, CP Holan’s and ˇ Zabokrtsk´ y ’s combining parser.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Prague Parsers

Prague Parsers

McD McDonnald’s maximum spanning tree parser, COL Collins’s parser adapted for PDT, ZZ ˇ Zabokrtsk´ y’s rule-based dependency parser, AN Holan’s parser ANALOG, L2R, R2L, L2R3, R2L3 Holan’s pushdown parsers, CP Holan’s and ˇ Zabokrtsk´ y ’s combining parser.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Prague Parsers

Prague Parsers

McD McDonnald’s maximum spanning tree parser, COL Collins’s parser adapted for PDT, ZZ ˇ Zabokrtsk´ y’s rule-based dependency parser, AN Holan’s parser ANALOG, L2R, R2L, L2R3, R2L3 Holan’s pushdown parsers, CP Holan’s and ˇ Zabokrtsk´ y ’s combining parser.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Prague Parsers

Prague Parsers

McD McDonnald’s maximum spanning tree parser, COL Collins’s parser adapted for PDT, ZZ ˇ Zabokrtsk´ y’s rule-based dependency parser, AN Holan’s parser ANALOG, L2R, R2L, L2R3, R2L3 Holan’s pushdown parsers, CP Holan’s and ˇ Zabokrtsk´ y ’s combining parser.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions synt Parser

synt Parser

Manually created meta-grammar (250 meta-rules), CF rules automatically generated (2800 CF rules), Semantic actions and contextual constraints, Head-driven chart parser + evaluation of the constraitns, More than 90 % of sentences from the PDTB-1.0 covered.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions synt Parser

synt Parser

Manually created meta-grammar (250 meta-rules), CF rules automatically generated (2800 CF rules), Semantic actions and contextual constraints, Head-driven chart parser + evaluation of the constraitns, More than 90 % of sentences from the PDTB-1.0 covered.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions synt Parser

synt Parser

Manually created meta-grammar (250 meta-rules), CF rules automatically generated (2800 CF rules), Semantic actions and contextual constraints, Head-driven chart parser + evaluation of the constraitns, More than 90 % of sentences from the PDTB-1.0 covered.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions synt Parser

synt Parser

Manually created meta-grammar (250 meta-rules), CF rules automatically generated (2800 CF rules), Semantic actions and contextual constraints, Head-driven chart parser + evaluation of the constraitns, More than 90 % of sentences from the PDTB-1.0 covered.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions synt Parser

synt Parser

Manually created meta-grammar (250 meta-rules), CF rules automatically generated (2800 CF rules), Semantic actions and contextual constraints, Head-driven chart parser + evaluation of the constraitns, More than 90 % of sentences from the PDTB-1.0 covered.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions

Differences of the Parses

1 Underlying formalism, 2 Input Format, 3 Dependency Trees vs. Phrasal Trees, 4 One Tree vs. (Shared) Forest, 5 Projective vs. Non-projective Trees, 6 Testing Data Set.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions

Differences of the Parses

1 Underlying formalism, 2 Input Format, 3 Dependency Trees vs. Phrasal Trees, 4 One Tree vs. (Shared) Forest, 5 Projective vs. Non-projective Trees, 6 Testing Data Set.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions

Differences of the Parses

1 Underlying formalism, 2 Input Format, 3 Dependency Trees vs. Phrasal Trees, 4 One Tree vs. (Shared) Forest, 5 Projective vs. Non-projective Trees, 6 Testing Data Set.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions

Differences of the Parses

1 Underlying formalism, 2 Input Format, 3 Dependency Trees vs. Phrasal Trees, 4 One Tree vs. (Shared) Forest, 5 Projective vs. Non-projective Trees, 6 Testing Data Set.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions

Differences of the Parses

1 Underlying formalism, 2 Input Format, 3 Dependency Trees vs. Phrasal Trees, 4 One Tree vs. (Shared) Forest, 5 Projective vs. Non-projective Trees, 6 Testing Data Set.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions

Differences of the Parses

1 Underlying formalism, 2 Input Format, 3 Dependency Trees vs. Phrasal Trees, 4 One Tree vs. (Shared) Forest, 5 Projective vs. Non-projective Trees, 6 Testing Data Set.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Input Format

Input Format

Morphological analyser (Ajka vs. Hajiˇ c’s tagger), Different morphological tagging systems.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Input Format

Input Format

Morphological analyser (Ajka vs. Hajiˇ c’s tagger), Different morphological tagging systems.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Dependency Trees vs. Phrasal Trees

Dependency Trees vs. Phrasal Trees

Dependecy Trees – The Prague Dependency Treebank No such large testing tree bank of phrasal trees Dependency → phrasal, Collins’s conversion tools

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Dependency Trees vs. Phrasal Trees

Dependency Trees vs. Phrasal Trees

Dependecy Trees – The Prague Dependency Treebank No such large testing tree bank of phrasal trees Dependency → phrasal, Collins’s conversion tools

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Dependency Trees vs. Phrasal Trees

Dependency Trees vs. Phrasal Trees

Dependecy Trees – The Prague Dependency Treebank No such large testing tree bank of phrasal trees Dependency → phrasal, Collins’s conversion tools

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions One Tree vs. (Shared) Forest

One Tree vs. (Shared) Forest

Dependency parsers – one tree at the output. synt – several trees at the output. First 100 trees (according to the tree rank) extracted, The best possible tree, the first tree, the average result of all 100.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions One Tree vs. (Shared) Forest

One Tree vs. (Shared) Forest

Dependency parsers – one tree at the output. synt – several trees at the output. First 100 trees (according to the tree rank) extracted, The best possible tree, the first tree, the average result of all 100.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions One Tree vs. (Shared) Forest

One Tree vs. (Shared) Forest

Dependency parsers – one tree at the output. synt – several trees at the output. First 100 trees (according to the tree rank) extracted, The best possible tree, the first tree, the average result of all 100.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions One Tree vs. (Shared) Forest

One Tree vs. (Shared) Forest

Dependency parsers – one tree at the output. synt – several trees at the output. First 100 trees (according to the tree rank) extracted, The best possible tree, the first tree, the average result of all 100.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Projective vs. Non-projective Trees

Projective vs. Non-projective Trees

Phrasal trees are not suitable to handle projectivity. 20 % of non-projective sentences in the PDT. Separate results for projective and non-projective sentences.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Projective vs. Non-projective Trees

Projective vs. Non-projective Trees

Phrasal trees are not suitable to handle projectivity. 20 % of non-projective sentences in the PDT. Separate results for projective and non-projective sentences.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Projective vs. Non-projective Trees

Projective vs. Non-projective Trees

Phrasal trees are not suitable to handle projectivity. 20 % of non-projective sentences in the PDT. Separate results for projective and non-projective sentences.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Measuring Techniques

Measuring Techniques

PARSEVAL crossing brackets Leaf-ancestor assessment (LAA) the sequence of node-labels found on the path between leaf and root nodes

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Measuring Techniques

Measuring Techniques

PARSEVAL crossing brackets Leaf-ancestor assessment (LAA) the sequence of node-labels found on the path between leaf and root nodes

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Measuring Techniques

Measuring Techniques

PARSEVAL crossing brackets Leaf-ancestor assessment (LAA) the sequence of node-labels found on the path between leaf and root nodes

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Results

Results – all sentences

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions Results

Results – small Brno treebank

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions

Conclusions and Future Directions

Conclusions Problems of the comparison were summarized, Prague stochastic parsers are better for general textual data, Comparable results on projective grammatical sentences. Current Development Larger treebank of Brno phrasal trees. Testing of dependencies generated by synt semantic actions.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions

Conclusions and Future Directions

Conclusions Problems of the comparison were summarized, Prague stochastic parsers are better for general textual data, Comparable results on projective grammatical sentences. Current Development Larger treebank of Brno phrasal trees. Testing of dependencies generated by synt semantic actions.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions

Conclusions and Future Directions

Conclusions Problems of the comparison were summarized, Prague stochastic parsers are better for general textual data, Comparable results on projective grammatical sentences. Current Development Larger treebank of Brno phrasal trees. Testing of dependencies generated by synt semantic actions.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions

Conclusions and Future Directions

Conclusions Problems of the comparison were summarized, Prague stochastic parsers are better for general textual data, Comparable results on projective grammatical sentences. Current Development Larger treebank of Brno phrasal trees. Testing of dependencies generated by synt semantic actions.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions

Conclusions and Future Directions

Conclusions Problems of the comparison were summarized, Prague stochastic parsers are better for general textual data, Comparable results on projective grammatical sentences. Current Development Larger treebank of Brno phrasal trees. Testing of dependencies generated by synt semantic actions.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison

Outline Motivation Compared Parsers Differences of the Parses Results Conclusions and Future Directions

Conclusions and Future Directions

Conclusions Problems of the comparison were summarized, Prague stochastic parsers are better for general textual data, Comparable results on projective grammatical sentences. Current Development Larger treebank of Brno phrasal trees. Testing of dependencies generated by synt semantic actions.

Aleˇ s Hor´ ak, Tom´ aˇ s Holan, Vladim´ ır Kadlec, Vojtˇ ech Kov´ aˇ r FI MU Brno and FMP CU Prague Dependency and Phrasal Parsers of the Czech Language: A Comparison