Upload
raine
View
28
Download
0
Tags:
Embed Size (px)
DESCRIPTION
Graph characterization of overlap-only TexMECS and other overlapping markup formalisms. Yves MARCOUX. GRDS – EBSI Université de Montréal. Visiting researcher Universitetet i Bergen. Overview of the talk. Graph representations of structured documents Overlap, in markup and structure - PowerPoint PPT Presentation
Citation preview
Yves Marcoux - Balisage 2008 1
Graph characterization of overlap-only TexMECS and other overlapping markup formalisms
Yves MARCOUX
GRDS – EBSIUniversité de Montréal
Visiting researcherUniversitetet i Bergen
Yves Marcoux - Balisage 2008 2
Overview of the talk
1. Graph representations of structured documents
2. Overlap, in markup and structure
3. OO-TexMECS (N.B.: Overlap-Only ;-)
4. Node-ordered DAGs (noDAGs)
5. Results and consequences
6. Future work
Yves Marcoux - Balisage 2008 3
1. Graph representations of structured documents
Yves Marcoux - Balisage 2008 4
XML document = tree
top
b
c
<top> <a> <b/> </a> <c/></top>
a
Embedding in markup Child-parent in tree
Yves Marcoux - Balisage 2008 5
Any tree an XML document
top
b
c
<top> <a> <b/> </a> <c/></top>
a
Yves Marcoux - Balisage 2008 6
Any tree an XML document
top
b
c
<top> <a> <b/> <d><e/></d> </a> <c/></top>
a
e
d
Perfect isomorphism !
Yves Marcoux - Balisage 2008 7
Document Object Models
• DOMs are essentially graph representations of structured documents
• "Patched" for attributes, namespaces, etc.
• DOM manipulations = graph modifications
• It suffices to make sure that the graph remains a tree
Yves Marcoux - Balisage 2008 8
2. Overlap, in markup and structure
Yves Marcoux - Balisage 2008 9
Problem of overlap
• In real life (outside of XML documents!), information is often not purely hierarchical
• Classical examples:– verse structure vs sentence structure– speech structure vs line structure
• In general: multiple structures applied (at least in part) to same contents
Yves Marcoux - Balisage 2008 10
Overlap
Yves Marcoux - Balisage 2008 11
Two views of overlap
• Geometric view: overlapping markup
• Common contents view: non-tree graph structure
Yves Marcoux - Balisage 2008 12
Example (markup)
(Peer) Hvorfor bande? (Åse) Tvi, du tør ej!Alt ihob er tøv og tant!
<vers> <peer>Hvorfor bande?</peer><åse>Tvi, du tør ej!</vers><vers> Alt ihob er tøv og tant!</åse></vers>
Yves Marcoux - Balisage 2008 13
Example (graph)
(Peer) Hvorfor bande? (Åse) Tvi, du tør ej!Alt ihob er tøv og tant!
vers vers
peeråse
Alt ihob er tøv og tant!Hvorfor bande? Tvi, du tør ej!
Yves Marcoux - Balisage 2008 14
Embedding Child-parent
vers vers
peeråse
Alt ihob er tøv og tant!Hvorfor bande? Tvi, du tør ej!
<vers> <peer>Hvorfor bande?</peer><åse>Tvi, du tør ej!</vers><vers> Alt ihob er tøv og tant!</åse></vers>
Yves Marcoux - Balisage 2008 15
Still perfect isomorphism?
• Not for graphs in general... cycles !
• Maybe for acyclic graphs ?
• Let's try more examples...
Yves Marcoux - Balisage 2008 16
What if the last verse is saidin chorus by Peer & Åse ?
vers vers
peeråse
Alt ihob er tøv og tant!Hvorfor bande? Tvi, du tør ej!
Yves Marcoux - Balisage 2008 17
What if the last verse is saidin chorus by Peer & Åse ?
vers vers
peer åse
Alt ihob er tøv og tant!Hvorfor bande? Tvi, du tør ej!
Still acyclic, but no corresponding markup !
Yves Marcoux - Balisage 2008 18
So, imperfect isomorphism
• Some acyclic graphs have corresponding documents (i.e., are serializable)
• Some (apparently) don't...
• Manipulations of the graph (DOM) may leave it non-serializable !
• Which acyclic graphs are serializable?
Yves Marcoux - Balisage 2008 19
3. OO-TexMECS
Yves Marcoux - Balisage 2008 20
TexMECS
• A particular proposal to address the overlap problem with overlapping markup
• MECS (Huitfeldt 1992-1996)– Multi-element code system
• TexMECS (Huitfeldt & SMcQ 2003)– "Trivially extended MECS"
• Markup Languages for Complex Documents (MLCD) project
Yves Marcoux - Balisage 2008 21
Overlap-only TexMECS
• TexMECS allows overlapping markup...
• but also much more:– virtual elements, interrupted elements, etc.
• OO-TexMECS 101– Start-tags: <a|– End-tags: |a>– Overlapping elements allowed– Natural notion of well-formedness
Yves Marcoux - Balisage 2008 22
4. Node-ordered DAGs (noDAGs)
Yves Marcoux - Balisage 2008 23
Contents ordering
• In a serialized document, contents appears in some order
• The order is often significant– Procedure steps, verses in a poem, etc.
• Thus, the order must be present in graph representations– XML: children ordered in the tree
Yves Marcoux - Balisage 2008 24
Node-ordered DAGs
• Node-ordered directed acyclic graphs
• Essentially a DAG with children ordered
• Why "node-ordered" (and not "child-ordered") ?– Provisions for further ordering of nodes
Yves Marcoux - Balisage 2008 25
Examples
A
B C
E F G H
D
B
C
D
E
A
Yves Marcoux - Balisage 2008 26
5. Results and consequences
Yves Marcoux - Balisage 2008 27
Main results
1. A noDAG is serializable in OO-TexMECS iff it is completion-acyclic (essentially)
2. Any OO-TexMECS well-formed document can be obtained by serializing some completion-acyclic noDAG
3. You don't gain any expressivity by allowing node-ordering over and above children-ordering
Yves Marcoux - Balisage 2008 28
Consequences
• We now know how to define a DOM for overlapping markup
• A DOM-based editor is complete
• Round-tripping is possible between noDAGs and OO-TexMECS
• Results also apply to similar formalisms– TexMECS with more features except virtual
and interrupted elements
Yves Marcoux - Balisage 2008 29
Completions
• Intuition: combination of parent-child relationships & children-ordering informs on the relative positioning of tags in an eventual serialization
• If contradictory information is derivable: the graph is not serializable
Yves Marcoux - Balisage 2008 30
Example: "starts-before"
A
B C
E F G H
D
Yves Marcoux - Balisage 2008 31
Example: "ends-after"
A
B C
E F G H
D
Yves Marcoux - Balisage 2008 32
Contradiction = cycle !
A
B C
E F G H
D
Completion-acyclic each completion is acyclic
Yves Marcoux - Balisage 2008 33
Future work
• Optimal algorithm for verification
• Optimal serialization algorithm
• Relaxing conditions on the graph
• Relationships with GODDAGs (SMcQ & Huitfeldt 2004)
• And, yes...
• Intertextual semantics of overlap !