Package edu.stanford.nlp.trees.tregex.tsurgeon

Class Tsurgeon

  • public class Tsurgeon
extends Object
    Tsurgeon provides a way of editing trees based on a set of operations that are applied to tree locations matching a tregex pattern. A simple example from the command-line:
    java edu.stanford.nlp.trees.tregex.tsurgeon.Tsurgeon -treeFile aTree exciseNP renameVerb
    The file aTree has Penn Treebank (S-expression) format trees. The other (here, two) files have Tsurgeon operations. These consist of a list of pairs of a tregex expression on one or more lines, a blank line, and then some number of lines of Tsurgeon operations and then another blank line.

    Tsurgeon uses the Tregex engine to match tree patterns on trees; for more information on Tregex's tree-matching functionality, syntax, and semantics, please see the documentation for the TregexPattern class.

    If you want to use Tsurgeon as an API, the relevant method is processPattern(edu.stanford.nlp.trees.tregex.TregexPattern, edu.stanford.nlp.trees.tregex.tsurgeon.TsurgeonPattern, edu.stanford.nlp.trees.Tree). You will also need to look at the TsurgeonPattern class and the parseOperation(java.lang.String) method.

    Here's the simplest form of invocation on a single Tree: 

     Tree t = Tree.valueOf("(ROOT (S (NP (NP (NNP Bank)) (PP (IN of) (NP (NNP America)))) (VP (VBD called)) (. .)))");
 TregexPattern pat = TregexPattern.compile("NP <1 (NP << Bank) <2 PP=remove");
 TsurgeonPattern surgery = Tsurgeon.parseOperation("excise remove remove");
 Tsurgeon.processPattern(pat, surgery, t).pennPrint();

    Here is another sample invocation: 

     TregexPattern matchPattern = TregexPattern.compile("SQ=sq < (/^WH/ $++ VP)");
 List<TsurgeonPattern> ps = new ArrayList<TsurgeonPattern>();

 TsurgeonPattern p = Tsurgeon.parseOperation("relabel sq S");

 ps.add(p);

 Treebank lTrees;
 List<Tree> result = Tsurgeon.processPatternOnTrees(matchPattern,Tsurgeon.collectOperations(ps),lTrees);

    Note: If you want to apply multiple surgery patterns, you will not want to call processPatternOnTrees, for each individual pattern. Rather, you should either call processPatternsOnTree and loop through the trees yourself, or, as above, use collectOperations to collect all the surgery patterns into one TsurgeonPattern, and then to call processPatternOnTrees. Either of these latter methods is much faster.

    The parser also has the ability to collect multiple TsurgeonPatterns into one pattern by itself by enclosing each pattern in [ ... ]. For example,
    Tsurgeon.parseOperation("[relabel foo BAR] [prune bar]")

    For more information on using Tsurgeon from the command line, see the main(java.lang.String[]) method and the package Javadoc.

    Author:
    Roger Levy

    • Method Detail

      • main

        public static void main​(String[] args)
                 throws Exception
        Usage: java edu.stanford.nlp.trees.tregex.tsurgeon.Tsurgeon [-s] -treeFile file-with-trees [-po matching-pattern operation] operation-file-1 operation-file-2 ... operation-file-n Arguments: Each argument should be the name of a transformation file that contains a list of pattern and transformation operation list pairs. That is, it is a sequence of pairs of a TregexPattern pattern on one or more lines, then a blank line (empty or whitespace), then a list of transformation operations one per line (as specified by Legal operation syntax below) to apply when the pattern is matched, and then another blank line (empty or whitespace). Note the need for blank lines: The code crashes if they are not present as separators (although the blank line at the end of the file can be omitted). The script file can include comment lines, either whole comment lines or trailing comments introduced by %, which extend to the end of line. A needed percent mark can be escaped by a preceding backslash.

        For example, if you want to excise an SBARQ node whenever it is the parent of an SQ node, and relabel the SQ node to S, your transformation file would look like this:

        SBARQ=n1 < SQ=n2

        excise n1 n1
        relabel n2 S
        Options:
        • -treeFile <filename> specify the name of the file that has the trees you want to transform.
        • -po <matchPattern> <operation> Apply a single operation to every tree using the specified match pattern and the specified operation. Use this option when you want to quickly try the effect of one pattern/surgery combination, and are too lazy to write a transformation file.
        • -s Print each output tree on one line (default is pretty-printing).
        • -m For every tree that had a matching pattern, print "before" (prepended as "Operated on:") and "after" (prepended as "Result:"). Unoperated on trees just pass through the transducer as usual.
        • -encoding X Uses character set X for input and output of trees.
        • -macros <filename> A file of macros to use on the tregex pattern. Macros should be one per line, with original and replacement separated by tabs.
        • -hf <headFinder-class-name> use the specified HeadFinder class to determine headship relations.
        • -hfArg <string> pass a string argument in to the HeadFinder class's constructor. -hfArg can be used multiple times to pass in multiple arguments.
        • -trf <TreeReaderFactory-class-name> use the specified TreeReaderFactory class to read trees from files.
        Legal operation syntax:
        • delete <name> deletes the node and everything below it.
        • prune <name> Like delete, but if, after the pruning, the parent has no children anymore, the parent is pruned too. Pruning continues to affect all ancestors until one is found with remaining children. This may result in a null tree.
        • excise <name1> <name2> The name1 node should either dominate or be the same as the name2 node. This excises out everything from name1 to name2. All the children of name2 go into the parent of name1, where name1 was.
        • relabel <name> <new-label> Relabels the node to have the new label.
          There are three possible forms:
          relabel nodeX VP - for changing a node label to an alphanumeric string
          relabel nodeX /<new-label>/ - for relabeling a node to something that isn't a valid identifier without quoting
          With this method, some replacement texts also require escaping. For example, relabel nodeX /{/ works but you need to do relabel nodeX /\\]/ in order to get a single close bracket. relabel nodeX /^VB(.*)$/verb\\/$1/ - for regular expression based relabeling. In this case, all matches of the regular expression against the node label are replaced with the replacement String. This has the semantics of Java/Perl's replaceAll: you may use capturing groups and put them in replacements with $n. For example, if the pattern is /foo/bar/ and the node matched is "foo", the replaceAll semantics result in "barbar". If the pattern is /^foo(.*)$/bar$1/ and node matched is "foofoo", relabel will result in "barfoo".
          When using the regex replacement method, you can also use the sequences ={node} and %{var} in the replacement string to use captured nodes or variable strings in the replacement string. For example, if the Tregex pattern was "duck=bar" and the relabel is /foo/={bar}/, "foofoo" will be replaced with "duckduck".
          To concatenate two nodes named in the tregex pattern, for example, you can use the pattern /^.*$/={foo}={bar}/. Note that the ^.*$ is necessary to make sure the regex pattern only matches and replaces once on the entire node name.
          To get an "=" or a "%" in the replacement, using \ escaping. Also, as in the example you can escape a slash in the middle of the second and third forms with \\/ and \\\\.
        • insert <name> <position> or insert <tree> <position> inserts the named node or tree into the position specified.
        • move <name> <position> moves the named node into the specified position.

          Right now the only ways to specify position are:

          $+ <name> the left sister of the named node
          $- <name> the right sister of the named node
          >i <name> the i_th daughter of the named node
          >-i <name> the i_th daughter, counting from the right, of the named node.

        • moveprune <name> <position> moves the named node into the specified position, then prunes the original position if it became a node with no children.
        • replace <name1> <name2> deletes name1 and inserts a copy of name2 in its place.
        • replace <name> <tree> <tree2>... deletes name and inserts the new tree(s) in its place. If more than one replacement tree is given, each of the new subtrees will be added in order where the old tree was. Multiple subtrees at the root is an illegal operation and will throw an exception.
        • createSubtree <auxiliary-tree-or-label> <name1> [<name2>] Create a subtree out of all the nodes from <name1> through <name2>. The subtree is moved to the foot of the given auxiliary tree, and the tree is inserted where the nodes of the subtree used to reside. If a simple label is provided as the first argument, the subtree is given a single parent with a name corresponding to the label. To limit the operation to just one node, elide <name2>.
        • adjoin <auxiliary_tree> <name> 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.
        • adjoinH <auxiliary_tree> <name> Similar to adjoin, but preserves the target node and makes it the root of <tree>. (It is still accessible as name. The root of the auxiliary tree is ignored.)
        • adjoinF <auxiliary_tree> <name> Similar to adjoin, but preserves the target node and makes it the foot of <tree>. (It is still accessible as name, and retains its status as parent of its children. The root of the auxiliary tree is ignored.)
        • coindex <name1> <name2> ... <nameM> Puts a (Penn Treebank style) coindexation suffix of the form "-N" on each of nodes name_1 through name_m. The value of N will be automatically generated in reference to the existing coindexations in the tree, so that there is never an accidental clash of indices across things that are not meant to be coindexed.

        In the context of adjoin, adjoinH, adjoinF, and createSubtree, an auxiliary tree is a tree in Penn Treebank format with @ on exactly one of the leaves denoting the foot of the tree. The operations which use the foot use the labeled node. For example:

        Tsurgeon: adjoin (FOO (BAR@)) foo
        Tregex: B=foo
        Input: (A (B 1 2))
        Output: (A (FOO (BAR 1 2)))

        Tsurgeon applies the same operation to the same tree for as long as the given tregex operation matches. This means that infinite loops are very easy to cause. One common situation where this comes up is with an insert operation will repeats infinitely many times unless you add an expression to the tregex that matches against the inserted pattern. For example, this pattern will infinite loop:

        TregexPattern tregex = TregexPattern.compile("S=node << NP");
        TsurgeonPattern tsurgeon = Tsurgeon.parseOperation("insert (NP foo) >-1 node");

        This pattern, though, will terminate:

        TregexPattern tregex = TregexPattern.compile("S=node << NP !<< foo");
        TsurgeonPattern tsurgeon = Tsurgeon.parseOperation("insert (NP foo) >-1 node");

        Tsurgeon has (very) limited support for conditional statements. If a pattern is prefaced with if exists <name>, the rest of the pattern will only execute if the named node was found in the corresponding TregexMatcher.

        Parameters:
        args - a list of names of files each of which contains a single tregex matching pattern plus a list, one per line, of transformation operations to apply to the matched pattern.
        Throws:
        Exception - If an I/O or pattern syntax error

      • getOperationFromReader

        public static Pair<TregexPattern,​TsurgeonPattern> getOperationFromReader​(BufferedReader reader,
                                                                               TregexPatternCompiler compiler)
                                                                        throws IOException
        Parses a tsurgeon script text input and compiles a tregex pattern and a list of tsurgeon operations into a pair.
        Parameters:
        reader - Reader to read patterns from
        Returns:
        A pair of a tregex and tsurgeon pattern read from a file, or null when the operations present in the Reader have been exhausted
        Throws:
        IOException - If any IO problem

      • getTregexPatternFromReader

        public static String getTregexPatternFromReader​(BufferedReader reader)
                                         throws IOException
        Assumes that we are at the beginning of a tsurgeon script file and gets the string for the tregex pattern leading the file.
        Returns:
        tregex pattern string. May be empty, never null
        Throws:
        IOException - If the usual kinds of IO errors occur

      • getTsurgeonOperationsFromReader

        public static TsurgeonPattern getTsurgeonOperationsFromReader​(BufferedReader reader)
                                                       throws IOException
        Assumes the given reader has only tsurgeon operations (not a tregex pattern), and parses these out, collecting them into one operation. Stops on a whitespace line.
        Throws:
        IOException - If the usual kinds of IO errors occur

      • getTsurgeonTextFromReader

        public static String getTsurgeonTextFromReader​(BufferedReader reader)
                                        throws IOException
        Assumes the given reader has only tsurgeon operations (not a tregex pattern), and returns them as a String, mirroring the way the strings appear in the file. This is helpful for lazy evaluation of the operations, as in a GUI, because you do not parse the operations on load. Comments are still excised.
        Throws:
        IOException

      • getOperationsFromFile

        public static List<Pair<TregexPattern,​TsurgeonPattern>> getOperationsFromFile​(String filename,
                                                                                    String encoding,
                                                                                    TregexPatternCompiler compiler)
                                                                             throws IOException
        Parses a tsurgeon script file and compiles all operations in the file into a list of pairs of tregex and tsurgeon patterns.
        Parameters:
        filename - A file, classpath resource or URL (perhaps gzipped) containing the tsurgeon script
        Returns:
        A pair of a tregex and tsurgeon pattern read from a file
        Throws:
        IOException - If there is any I/O problem

      • getOperationsFromReader

        public static List<Pair<TregexPattern,​TsurgeonPattern>> getOperationsFromReader​(BufferedReader reader,
                                                                                      TregexPatternCompiler compiler)
                                                                               throws IOException
        Parses and compiles all operations from a BufferedReader into a list of pairs of tregex and tsurgeon patterns.
        Parameters:
        reader - A BufferedReader to read the operations
        Returns:
        A pair of a tregex and tsurgeon pattern read from reader
        Throws:
        IOException - If there is any I/O problem

      • processPattern

        public static Tree processPattern​(TregexPattern matchPattern,
                                  TsurgeonPattern p,
                                  Tree t)
        Tries to match a pattern against a tree. If it succeeds, apply the surgical operations contained in a TsurgeonPattern.
        Parameters:
        matchPattern - A TregexPattern to be matched against a Tree.
        p - A TsurgeonPattern to apply.
        t - the Tree to match against and perform surgery on.
        Returns:
        t, which has been surgically modified.

      • parseOperation

        public static TsurgeonPattern parseOperation​(String operationString)
        Parses an operation string into a TsurgeonPattern. Throws an TsurgeonParseException if the operation string is ill-formed.

        Example of use:

        TsurgeonPattern p = Tsurgeon.parseOperation("prune ed");

        Parameters:
        operationString - The operation to perform, as a text string
        Returns:
        the operation pattern.

      • collectOperations

        public static TsurgeonPattern collectOperations​(List<TsurgeonPattern> patterns)
        Collects a list of operation patterns into a sequence of operations to be applied. Required to keep track of global properties across a sequence of operations. For example, if you want to insert a named node and then coindex it with another node, you will need to collect the insertion and coindexation operations into a single TsurgeonPattern so that tsurgeon is aware of the name of the new node and coindexation becomes possible.
        Parameters:
        patterns - a list of TsurgeonPattern operations that you want to collect together into a single compound operation
        Returns:
        a new TsurgeonPattern that performs all the operations in the sequence of the patterns argument