How to perform tree surgery

1
How to perform tree surgery

• Anna Rafferty
• Marie-Catherine de Marneffe

2
Tsurgeon by Roger Levy

• What?
• makes operations on a grammatical tree
• How?
• based on Tregex syntax
• Where?
• Javanlp trees.tregex.tsurgeon

3
How? Tregex

• utility for identifying patterns in trees
• (like regular expressions for strings)
• node descriptions and relationships between
  nodes

NP lt /NN/
4
Tsurgeon syntax

• Define a pattern to be matched on the trees
• VBZvbz NP
• Define one or several operation(s)
• relabel vbz VBZ_TRANSITIVE

5
Delete

• (ROOT
• (SBARQ
• (SQ (NP (NNS Cats))
• (VP (VBP do)
• (VP (WHNP what)
• (VB eat)))
• (PUNCT ?)))
• PUNCTpunct gt SBARQ
• delete punct

6
Delete

• (ROOT
• (SBARQ
• (SQ (NP (NNS Cats))
• (VP (VBP do)
• (VP (WHNP what)
• (VB eat)))
• (PUNCT ?)))
• PUNCTpunct gt SBARQ
• delete punct

delete ltname1gtltnameNgt
Delete the node and everything below it
7
Excise

• (ROOT
• (SBARQ
• (SQ (NP (NNS Cats))
• (VP (VBP do)
• (VP (WHNP what)
• (VB eat))))))
• SBARQsbarq gt ROOT
• excise sbarq sbarq

(ROOT (SQ (NP (NNS Cats)) (VP (VBP
do) (VP (WHNP what) (VB
eat)))))
8
Excise

• (ROOT
• (SBARQ
• (SQ (NP (NNS Cats))
• (VP (VBP do)
• (VP (WHNP what)
• (VB eat))))))
• SBARQsbarq gt ROOT
• excise sbarq sbarq

excise ltname1gt ltname2gt
name1 is name2 or dominates name2. All children
of name2 go into the parent of name1, where
name1 was.
9
Prune

• prune ltname1gtltnameNgt
• Different from delete
• If after the pruning the parent has no children
• anymore, the parent is pruned too.

10
Insert

• (ROOT
• (SQ (NP (NNS Cats))
• (VP (VBP do)
• (VP (WHNP what)
• (VB eat)))))
• SQsq gt ROOT !lt- /PUNCT/
• insert (PUNCT .) gt-1 sq
• lttreegt ltpositiongt

(ROOT (SQ (NP (NNS Cats)) (VP (VBP
do) (VP (WHNP what) (VB
eat))) (PUNCT .)))
Caveat cyclic application of rules
11
Position for insert and move

• insert ltnamegt ltpositiongt
• insert lttreegt ltpositiongt
• ltpositiongt ltrelationgt ltnamegt
• ltrelationgt
• the left sister of the named node
• - the right sister of the named node
• gti the i_th daughter of the named node
• gt-i the i_th daughter, counting from the
  right, of the named node.

12
Move

• (ROOT
• (SQ
• (NP (NNS Cats))
• (VP (VBP do)
• (VP (WHNP what)
• (VB eat)))
• (PUNCT .)))
• VP lt (/WH/wh /VB/vb)
• move vb wh
• ltpositiongt

move ltnamegt ltpositiongt
moves the named node into the specified position
13
Move

• (ROOT
• (SQ
• (NP (NNS Cats))
• (VP (VBP do)
• (VP (WHNP what)
• (VB eat)))
• (PUNCT .)))
• VP lt (/WH/wh /VB/vb)
• move vb wh
• ltpositiongt

(ROOT (SQ (NP (NNS Cats)) (VP
(VBP do) (VP (VB eat) (WHNP
what))) (PUNCT .)))
14
Adjoin
(ROOT (SQ (NP (NNS Cats)) (VP
(ADVP (RB usually)) (VP (VBP do)
(VP (VB eat) (WHNP
what))) (PUNCT .)))

• (ROOT
• (SQ
• (NP (NNS Cats))
• (VP (VBP do)
• (VP (VB eat)
• (WHNP what)))
• (PUNCT .)))

VPvp gt SQ !gt (__ ltlt usually) adjoin (VP (ADVP
(ADV usually)) VP_at_) vp
15
Adjoin syntax

• adjoin ltauxiliary_treegt ltnamegt
• Adjoins the specified auxiliary tree into the
• named node.
• The daughters of the target node will become
• the daughters of the foot of the auxiliary tree.
• adjoin (VP (ADVP (ADV usually)) VP_at_) vp
• foot

16
On the command line

• java Tsurgeon -treeFile
• ltaFilegt ltoperationFilegt
• aFile -gt a file containing the trees to be
  transformed
• operationFile -gt pattern (Tregex expression)
• an empty line
• operation(s) (one by line)

17
How to use the Tsurgeon class

• TregexPattern matchPattern TregexPattern.compile
  ("SQsq lt (/WH/ VP)")
• ListltTsurgeonPatterngt ps
• new ArrayListltTsurgeonPatterngt()
• TsurgeonPattern p
• Tsurgeon.parseOperation("relabel sq S")
• ps.add(p)
• CollectionltTreegt result Tsurgeon.processPatternO
  nTrees(matchPattern,Tsurgeon.collectOperations(ps)
  ,lTrees)

18
To become a specialist

• See Rogers README!
• Practice tree surgery!