Graph Parsing (State of the Art) - Revision history

DanielH: /* UCCA: Universal Conceptual Cognitive Annotation */

2017-06-01T15:13:51Z

UCCA: Universal Conceptual Cognitive Annotation

← Older revision		Revision as of 15:13, 1 June 2017
Line 192:		Line 192:
	\|average number of tokens		\|average number of tokens
	\|style="text-align:right;"\|431.7		\|style="text-align:right;"\|431.7
	\|style="text-align:right;"\|		\|style="text-align:right;"\|30.32
	\|style="text-align:right;"\|		\|style="text-align:right;"\|114.24
	\|style="text-align:right;"\|		\|style="text-align:right;"\|34.77
	\|-		\|-
	\|average number of nodes per token		\|average number of nodes per token
	\|style="text-align:right;"\|		\|style="text-align:right;"\|1.89
	\|style="text-align:right;"\|		\|style="text-align:right;"\|1.89
	\|style="text-align:right;"\|		\|style="text-align:right;"\|1.67
	\|style="text-align:right;"\|		\|style="text-align:right;"\|1.67
	\|-		\|-
	\|number of edge labels		\|number of edge labels

DanielH: /* UCCA: Universal Conceptual Cognitive Annotation */

2017-06-01T15:09:21Z

UCCA: Universal Conceptual Cognitive Annotation

← Older revision		Revision as of 15:09, 1 June 2017
Line 192:		Line 192:
	\|average number of tokens		\|average number of tokens
	\|style="text-align:right;"\|431.7		\|style="text-align:right;"\|431.7
			\|style="text-align:right;"\|
			\|style="text-align:right;"\|
			\|style="text-align:right;"\|
	\|-		\|-
	\|average number of nodes per token		\|average number of nodes per token
			\|style="text-align:right;"\|
			\|style="text-align:right;"\|
			\|style="text-align:right;"\|
	\|style="text-align:right;"\|		\|style="text-align:right;"\|
	\|-		\|-
Line 203:		Line 209:
	\|-		\|-
	\|percentage of graphs that are trees		\|percentage of graphs that are trees
			\|style="text-align:right;"\|
			\|style="text-align:right;"\|
			\|style="text-align:right;"\|
	\|style="text-align:right;"\|		\|style="text-align:right;"\|
	\|-		\|-
	\|average edge density		\|average edge density
			\|style="text-align:right;"\|
			\|style="text-align:right;"\|
			\|style="text-align:right;"\|
	\|style="text-align:right;"\|		\|style="text-align:right;"\|
	\|-		\|-
	\|percentage of nodes that are reentrant		\|percentage of nodes that are reentrant
			\|style="text-align:right;"\|
			\|style="text-align:right;"\|
			\|style="text-align:right;"\|
	\|style="text-align:right;"\|		\|style="text-align:right;"\|
	\|-		\|-
Line 230:		Line 245:
	\|-		\|-
	\|percentage of non-top roots		\|percentage of non-top roots
			\|style="text-align:right;"\|
			\|style="text-align:right;"\|
			\|style="text-align:right;"\|
	\|style="text-align:right;"\|		\|style="text-align:right;"\|
	\|-		\|-
	\|percentage of graphs that are noncrossing		\|percentage of graphs that are noncrossing
			\|style="text-align:right;"\|
			\|style="text-align:right;"\|
			\|style="text-align:right;"\|
	\|style="text-align:right;"\|		\|style="text-align:right;"\|
	\|}		\|}

	= References =		= References =

DanielH: /* UCCA: Universal Conceptual Cognitive Annotation */

2017-06-01T15:08:28Z

UCCA: Universal Conceptual Cognitive Annotation

← Older revision		Revision as of 15:08, 1 June 2017
Line 197:		Line 197:
	\|-		\|-
	\|number of edge labels		\|number of edge labels
	\|style="text-align:right;"\|12\|12\|12\|12		\|style="text-align:right;"\|12
			\|style="text-align:right;"\|12
			\|style="text-align:right;"\|12
			\|style="text-align:right;"\|12
	\|-		\|-
	\|percentage of graphs that are trees		\|percentage of graphs that are trees
Line 209:		Line 212:
	\|-		\|-
	\|percentage of graphs that are cyclic		\|percentage of graphs that are cyclic
	\|style="text-align:right;"\|0\|0\|0\|0		\|style="text-align:right;"\|0
			\|style="text-align:right;"\|0
			\|style="text-align:right;"\|0
			\|style="text-align:right;"\|0
	\|-		\|-
	\|percentage of graphs that are not connected		\|percentage of graphs that are not connected
	\|style="text-align:right;"\|0\|0\|0\|0		\|style="text-align:right;"\|0
			\|style="text-align:right;"\|0
			\|style="text-align:right;"\|0
			\|style="text-align:right;"\|0
	\|-		\|-
	\|percentage of graphs that are multi-rooted		\|percentage of graphs that are multi-rooted
	\|style="text-align:right;"\|0\|0\|0\|0		\|style="text-align:right;"\|0
			\|style="text-align:right;"\|0
			\|style="text-align:right;"\|0
			\|style="text-align:right;"\|0
	\|-		\|-
	\|percentage of non-top roots		\|percentage of non-top roots

DanielH: /* UCCA: Universal Conceptual Cognitive Annotation */

2017-06-01T15:07:48Z

UCCA: Universal Conceptual Cognitive Annotation

← Older revision		Revision as of 15:07, 1 June 2017
Line 179:		Line 179:
	{\|		{\|
	! style="text-align:left;"\| Property		! style="text-align:left;"\| Property
	! style="text-align:center;"\| UCCA Wiki \| UCCA Wiki (sentences) \| UCCA 20K Leagues \| UCCA 20K Leagues (sentences)		! style="text-align:center;"\| UCCA Wiki
			! style="text-align:center;" \| UCCA Wiki (sentences)
			! style="text-align:center;"\| UCCA 20K Leagues
			! style="text-align:center;"\| UCCA 20K Leagues (sentences)
	\|-		\|-
	\|number of graphs		\|number of graphs
	\|style="text-align:right;"\|367\|5225\|154\|506		\|style="text-align:right;"\|367
			\|style="text-align:right;"\|5225
			\|style="text-align:right;"\|154
			\|style="text-align:right;"\|506
	\|-		\|-
	\|average number of tokens		\|average number of tokens
Line 217:		Line 223:
	\|style="text-align:right;"\|		\|style="text-align:right;"\|
	\|}		\|}


	= References =		= References =

DanielH: Add some UCCA details

2017-06-01T15:06:25Z

Add some UCCA details

← Older revision		Revision as of 15:06, 1 June 2017
Line 170:		Line 170:

	= UCCA: Universal Conceptual Cognitive Annotation =		= UCCA: Universal Conceptual Cognitive Annotation =
			[http://www.cs.huji.ac.il/~oabend/ucca.html Universal Conceptual Cognitive Annotation (UCCA)] is a novel semantic approach to grammatical representation. It was developed in the Computational Linguistics Lab of the Computer Science Department of the Hebrew University by Omri Abend and Ari Rappoport<ref name="abend2013">Abend, Omri and Ari Rappoport. 2013. Universal Conceptual Cognitive Annotation (UCCA). ACL.</ref>.

			The central idea of the project is to analyze and annotate natural languages using purely semantic categories and structure (a graph). Syntactic categories and structure are automatically deduced from the semantic ones using learning algorithms. The basic set of semantic categories (the foundational layer) is inspired by work in linguistic typology, cognitive grammar, and neuroscience.

			Hershcovich et al. introduced the first parser for UCCA <ref name="hershcovich2017">Hershcovich, Daniel and Omri Abend and Ari Rappoport. 2017. A Transition-Based Directed Acyclic Graph Parser for UCCA. ACL.</ref>, a transition-based parser based on BiLSTM feature representation.


			{\|
			! style="text-align:left;"\| Property
			! style="text-align:center;"\| UCCA Wiki \| UCCA Wiki (sentences) \| UCCA 20K Leagues \| UCCA 20K Leagues (sentences)
			\|-
			\|number of graphs
			\|style="text-align:right;"\|367\|5225\|154\|506
			\|-
			\|average number of tokens
			\|style="text-align:right;"\|431.7
			\|-
			\|average number of nodes per token
			\|style="text-align:right;"\|
			\|-
			\|number of edge labels
			\|style="text-align:right;"\|12\|12\|12\|12
			\|-
			\|percentage of graphs that are trees
			\|style="text-align:right;"\|
			\|-
			\|average edge density
			\|style="text-align:right;"\|
			\|-
			\|percentage of nodes that are reentrant
			\|style="text-align:right;"\|
			\|-
			\|percentage of graphs that are cyclic
			\|style="text-align:right;"\|0\|0\|0\|0
			\|-
			\|percentage of graphs that are not connected
			\|style="text-align:right;"\|0\|0\|0\|0
			\|-
			\|percentage of graphs that are multi-rooted
			\|style="text-align:right;"\|0\|0\|0\|0
			\|-
			\|percentage of non-top roots
			\|style="text-align:right;"\|
			\|-
			\|percentage of graphs that are noncrossing
			\|style="text-align:right;"\|
			\|}


	= References =		= References =

Oe: /* SDP: Semantic Dependency Parsing */

2016-06-07T20:45:51Z

SDP: Semantic Dependency Parsing

← Older revision		Revision as of 20:45, 7 June 2016
Line 168:		Line 168:

	= SDP: Semantic Dependency Parsing =		= SDP: Semantic Dependency Parsing =

			= UCCA: Universal Conceptual Cognitive Annotation =

	= References =		= References =

MarcoKuhlmann at 09:31, 30 May 2016

2016-05-30T09:31:20Z

← Older revision		Revision as of 09:31, 30 May 2016
Line 40:		Line 40:
	= AMR: Abstract Meaning Representation =		= AMR: Abstract Meaning Representation =

	Abstract Meaning Representation (AMR) eschews explicit syntactic derivations and consideration of the syntax–semantics interface; it rather seeks to directly annotate “whole-sentence logical meanings” (Banarescu et al. 2013<ref name="banarescu2013">Banarescu, Laura, Claire Bonial, Shu Cai, Madalina Georgescu, Kira Griffitt, Ulf Hermjakob, Kevin Knight, Philipp Koehn, Martha Palmer, and Nathan Schneider. 2013. Abstract Meaning Representation for sembanking. In Proceedings of the 7th Linguistic Annotation Workshop and Interoperability with Discourse, ~~page~~ 178–186, Sofia, Bulgaria, August.</ref>). Node labels in AMR name abstract concepts, which in large parts draw on the ontology of OntoNotes predicate senses and corresponding semantic roles. Nodes are not overtly related to surface lexical units, and thus are unordered. Although AMR has its roots in semantic networks and earlier knowledge representation approaches (Langkilde and Knight 1998<ref name="langkilde1998">Langkilde, Irene and Kevin Knight. 1998. Generation that exploits corpus-based statistical		Abstract Meaning Representation (AMR) eschews explicit syntactic derivations and consideration of the syntax–semantics interface; it rather seeks to directly annotate “whole-sentence logical meanings” (Banarescu et al. 2013<ref name="banarescu2013">Banarescu, Laura, Claire Bonial, Shu Cai, Madalina Georgescu, Kira Griffitt, Ulf Hermjakob, Kevin Knight, Philipp Koehn, Martha Palmer, and Nathan Schneider. 2013. Abstract Meaning Representation for sembanking. In Proceedings of the 7th Linguistic Annotation Workshop and Interoperability with Discourse, pages 178–186, Sofia, Bulgaria, August.</ref>). Node labels in AMR name abstract concepts, which in large parts draw on the ontology of OntoNotes predicate senses and corresponding semantic roles. Nodes are not overtly related to surface lexical units, and thus are unordered. Although AMR has its roots in semantic networks and earlier knowledge representation approaches (Langkilde and Knight 1998<ref name="langkilde1998">Langkilde, Irene and Kevin Knight. 1998. Generation that exploits corpus-based statistical
	knowledge. In Proceedings of the 17th International Conference on Computational Linguistics and the 36th Meeting of the Association for Computational Linguistics, ~~page~~ 704–710, Montréal, Canada.</ref>), larger-scale manual AMR annotation is a recent development only. We sample two variants of AMR, viz. (a) the graphs as annotated in AMRBank 1.0 (LDC2014T12), and (b) a normalized version that we call AMR<sup>−1</sup>, where so-called “inverse roles” (like ARG0-of) are reversed. Such inverted edges are frequently used in AMR in order to render the graph as a single rooted structure, where the root is interpreted as the top-level focus. (The graph bank is natively constructed and released with inverted edges, but for parser evaluation the AMR<sup>−1</sup> normalization is typically assumed; our conversion builds on the code of Cai and Knight (2013)<ref name="cai2013">Cai, Shu and Kevin Knight. 2013. Smatch. An evaluation metric for semantic feature structures. In Proceedings of the 51th Meeting of the Association for Computational Linguistics, ~~page~~ 748–752, Sofia, Bulgaria, August.</ref>.) In the context of this comparison, we map this interpretation to a general concept of “top nodes” for both AMR and AMR<sup>−1</sup>. Flanigan et al. (2014)<ref name="flanigan2014">Flanigan, Jeffrey, Sam Thomson, Jaime Carbonell, Chris Dyer, and Noah A. Smith. 2014. A discriminative graph-based parser for the Abstract Meaning Representation. In Proceedings of the 52nd Meeting of the Association for Computational Linguistics, ~~page~~ 1426–1436, Baltimore, MD, USA, June.</ref> published the first parser targeting AMR, and the state of the art has been repeatedly updated since.		knowledge. In Proceedings of the 17th International Conference on Computational Linguistics and the 36th Meeting of the Association for Computational Linguistics, pages 704–710, Montréal, Canada.</ref>), larger-scale manual AMR annotation is a recent development only. We sample two variants of AMR, viz. (a) the graphs as annotated in AMRBank 1.0 (LDC2014T12), and (b) a normalized version that we call AMR<sup>−1</sup>, where so-called “inverse roles” (like ARG0-of) are reversed. Such inverted edges are frequently used in AMR in order to render the graph as a single rooted structure, where the root is interpreted as the top-level focus. (The graph bank is natively constructed and released with inverted edges, but for parser evaluation the AMR<sup>−1</sup> normalization is typically assumed; our conversion builds on the code of Cai and Knight (2013)<ref name="cai2013">Cai, Shu and Kevin Knight. 2013. Smatch. An evaluation metric for semantic feature structures. In Proceedings of the 51th Meeting of the Association for Computational Linguistics, pages 748–752, Sofia, Bulgaria, August.</ref>.) In the context of this comparison, we map this interpretation to a general concept of “top nodes” for both AMR and AMR<sup>−1</sup>. Flanigan et al. (2014)<ref name="flanigan2014">Flanigan, Jeffrey, Sam Thomson, Jaime Carbonell, Chris Dyer, and Noah A. Smith. 2014. A discriminative graph-based parser for the Abstract Meaning Representation. In Proceedings of the 52nd Meeting of the Association for Computational Linguistics, pages 1426–1436, Baltimore, MD, USA, June.</ref> published the first parser targeting AMR, and the state of the art has been repeatedly updated since.

	{\|		{\|
Line 107:		Line 107:
	= CCD: Combinatory Categorial Grammar Dependencies =		= CCD: Combinatory Categorial Grammar Dependencies =

	Hockenmaier and Steedman (2007)<ref name="hockenmaier2007">Hockenmaier, Julia and Mark Steedman. 2007. CCGbank. A corpus of CCG derivations and dependency structures extracted from the Penn Treebank. Computational Linguistics, 33:355–396.</ref> construct CCGbank from a combination of careful interpretation of the syntactic annotations in the PTB with additional, manually curated lexical and constructional knowledge. In CCGbank (LDC2005T13), the strings of the venerable PTB Wall Street Journal (WSJ) corpus are annotated with pairs of (a) CCG syntactic derivations and (b) sets of semantic bi-lexical dependency triples, which we term CCD. The latter “include most semantically relevant non-anaphoric local and long-range dependencies” and are suggested by the CCGbank creators as a proxy for predicate–argument structure. While CCD has mainly been used for contrastive parser evaluation (Clark and Curran [2007]<ref name="clark2007">Clark, Stephen and James R. Curran. 2007. Wide-coverage efficient statistical parsing with CCG and log-linear models. Computational Linguistics, 33(4):493–552.</ref>, Fowler and Penn [2010]<ref name="fowler2010">Fowler, Timothy A. D. and Gerald Penn. 2010. Accurate context-free parsing with Combinatory Categorial Grammar. In Proceedings of the 48th Meeting of the Association for Computational Linguistics, ~~page~~ 335–344, Uppsala, Sweden.</ref>; inter alios), there is current work that views each set of triples as a directed graph and parses directly into these target representations (Du, Sun, and Wan 2015<ref name="du2015">Du, Yantao, Weiwei Sun, and Xiaojun Wan. 2015. A data-driven, factorization parser for CCG dependency structures. In Proceedings of the 53rd Meeting of the Association for Computational Linguistics and of the 7th International Joint Conference on Natural Language Processing, ~~page~~ 1545–1555, Bejing, China.</ref>).		Hockenmaier and Steedman (2007)<ref name="hockenmaier2007">Hockenmaier, Julia and Mark Steedman. 2007. CCGbank. A corpus of CCG derivations and dependency structures extracted from the Penn Treebank. Computational Linguistics, 33:355–396.</ref> construct CCGbank from a combination of careful interpretation of the syntactic annotations in the PTB with additional, manually curated lexical and constructional knowledge. In CCGbank (LDC2005T13), the strings of the venerable PTB Wall Street Journal (WSJ) corpus are annotated with pairs of (a) CCG syntactic derivations and (b) sets of semantic bi-lexical dependency triples, which we term CCD. The latter “include most semantically relevant non-anaphoric local and long-range dependencies” and are suggested by the CCGbank creators as a proxy for predicate–argument structure. While CCD has mainly been used for contrastive parser evaluation (Clark and Curran [2007]<ref name="clark2007">Clark, Stephen and James R. Curran. 2007. Wide-coverage efficient statistical parsing with CCG and log-linear models. Computational Linguistics, 33(4):493–552.</ref>, Fowler and Penn [2010]<ref name="fowler2010">Fowler, Timothy A. D. and Gerald Penn. 2010. Accurate context-free parsing with Combinatory Categorial Grammar. In Proceedings of the 48th Meeting of the Association for Computational Linguistics, pages 335–344, Uppsala, Sweden.</ref>; inter alios), there is current work that views each set of triples as a directed graph and parses directly into these target representations (Du, Sun, and Wan 2015<ref name="du2015">Du, Yantao, Weiwei Sun, and Xiaojun Wan. 2015. A data-driven, factorization parser for CCG dependency structures. In Proceedings of the 53rd Meeting of the Association for Computational Linguistics and of the 7th International Joint Conference on Natural Language Processing, pages 1545–1555, Bejing, China.</ref>).

	{\|		{\|

Oe: /* SDP: Semantic Dependency Parsing */

2016-05-30T09:26:18Z

SDP: Semantic Dependency Parsing

← Older revision		Revision as of 09:26, 30 May 2016
Line 168:		Line 168:

	= SDP: Semantic Dependency Parsing =		= SDP: Semantic Dependency Parsing =

			= References =

MarcoKuhlmann at 07:12, 30 May 2016

2016-05-30T07:12:18Z

← Older revision		Revision as of 07:12, 30 May 2016
Line 24:		Line 24:
	state-of-the-art processing results.		state-of-the-art processing results.
	Of these, components (a) and (b) are provided by		Of these, components (a) and (b) are provided by
	[http://www.mn.uio.no/ifi/english/people/aca/oe/cl.pdf ~~Kuhlmann & Oepen (2016; in press)~~],		Kuhlmann and Oepen (2016)<ref name="kuhlmann2016">Marco Kuhlmann and Stephan Oepen. Towards a Catalogue of Linguistic Graph Banks. Computational Linguistics, 2016. In press. [http://www.mn.uio.no/ifi/english/people/aca/oe/cl.pdf Preprint]</ref>,
	while (c) and (d) are maintained below.		while (c) and (d) are maintained below.

Line 35:		Line 35:

	= Software: Graph Analysis Toolkit =		= Software: Graph Analysis Toolkit =

			An open-source reference implementation of the toolkit that was built to conduct the study of Kuhlmann and Oepen (2016)<ref name="kuhlmann2016"/> will be available later this summer.

	= AMR: Abstract Meaning Representation =		= AMR: Abstract Meaning Representation =

MarcoKuhlmann at 07:04, 30 May 2016

2016-05-30T07:04:35Z

← Older revision		Revision as of 07:04, 30 May 2016
Line 105:		Line 105:
	= CCD: Combinatory Categorial Grammar Dependencies =		= CCD: Combinatory Categorial Grammar Dependencies =

	Hockenmaier and Steedman (2007)<ref name="hockenmaier2007">Hockenmaier, Julia and Mark Steedman. 2007. CCGbank. A corpus of CCG derivations and dependency structures extracted from the Penn Treebank. Computational Linguistics, 33:355–396.</ref> construct CCGbank from a combination of careful interpretation of the syntactic annotations in the PTB with additional, manually curated lexical and constructional knowledge. In CCGbank (LDC2005T13), the strings of the venerable PTB Wall Street Journal (WSJ) corpus are annotated with pairs of (a) CCG syntactic derivations and (b) sets of semantic bi-lexical dependency triples, which we term CCD. The latter “include most semantically relevant non-anaphoric local and long-range dependencies” and are suggested by the CCGbank creators as a proxy for predicate–argument structure. While CCD has mainly been used for contrastive parser evaluation (Clark and Curran [2007]<ref name="clark2007">Clark, Stephen and James R. Curran. 2007. Wide-coverage efficient statistical parsing with CCG and log-linear models. Computational Linguistics, 33(4):493–552.</ref>, Fowler and Penn [2010]<ref name="fowler2010">Fowler, Timothy A. D. and Gerald Penn. 2010. Accurate context-free parsing with Combinatory Categorial Grammar. In Proceedings of the 48th Meeting of the Association for Computational Linguistics, page 335–344, Uppsala, Sweden.</ref>; inter alios), there is current work that views each set of triples as a directed graph and parses directly into these target representations (Du, Sun, and Wan 2015<ref="du2015">Du, Yantao, Weiwei Sun, and Xiaojun Wan. 2015. A data-driven, factorization parser for CCG dependency structures. In Proceedings of the 53rd Meeting of the Association for Computational Linguistics and of the 7th International Joint Conference on Natural Language Processing, page 1545–1555, Bejing, China.</ref>).		Hockenmaier and Steedman (2007)<ref name="hockenmaier2007">Hockenmaier, Julia and Mark Steedman. 2007. CCGbank. A corpus of CCG derivations and dependency structures extracted from the Penn Treebank. Computational Linguistics, 33:355–396.</ref> construct CCGbank from a combination of careful interpretation of the syntactic annotations in the PTB with additional, manually curated lexical and constructional knowledge. In CCGbank (LDC2005T13), the strings of the venerable PTB Wall Street Journal (WSJ) corpus are annotated with pairs of (a) CCG syntactic derivations and (b) sets of semantic bi-lexical dependency triples, which we term CCD. The latter “include most semantically relevant non-anaphoric local and long-range dependencies” and are suggested by the CCGbank creators as a proxy for predicate–argument structure. While CCD has mainly been used for contrastive parser evaluation (Clark and Curran [2007]<ref name="clark2007">Clark, Stephen and James R. Curran. 2007. Wide-coverage efficient statistical parsing with CCG and log-linear models. Computational Linguistics, 33(4):493–552.</ref>, Fowler and Penn [2010]<ref name="fowler2010">Fowler, Timothy A. D. and Gerald Penn. 2010. Accurate context-free parsing with Combinatory Categorial Grammar. In Proceedings of the 48th Meeting of the Association for Computational Linguistics, page 335–344, Uppsala, Sweden.</ref>; inter alios), there is current work that views each set of triples as a directed graph and parses directly into these target representations (Du, Sun, and Wan 2015<ref name="du2015">Du, Yantao, Weiwei Sun, and Xiaojun Wan. 2015. A data-driven, factorization parser for CCG dependency structures. In Proceedings of the 53rd Meeting of the Association for Computational Linguistics and of the 7th International Joint Conference on Natural Language Processing, page 1545–1555, Bejing, China.</ref>).

	{\|		{\|

← Older revision		Revision as of 15:13, 1 June 2017
Line 192:		Line 192:
	\|average number of tokens		\|average number of tokens
	\|style="text-align:right;"\|431.7		\|style="text-align:right;"\|431.7
	\|style="text-align:right;"\|		\|style="text-align:right;"\|30.32
	\|style="text-align:right;"\|		\|style="text-align:right;"\|114.24
	\|style="text-align:right;"\|		\|style="text-align:right;"\|34.77
	\|-		\|-
	\|average number of nodes per token		\|average number of nodes per token
	\|style="text-align:right;"\|		\|style="text-align:right;"\|1.89
	\|style="text-align:right;"\|		\|style="text-align:right;"\|1.89
	\|style="text-align:right;"\|		\|style="text-align:right;"\|1.67
	\|style="text-align:right;"\|		\|style="text-align:right;"\|1.67
	\|-		\|-
	\|number of edge labels		\|number of edge labels