Link Search Menu Expand Document
TGrep2 and TDT

TGrep2

Table of contents

  1. Basics
  2. Tree search

In this tutorial, you’ll walk through some simple tgrep2 commands to become familiar with the basic options and basic tree search patterns.

Basics

General usage:

tgrep2 [options] <pattern>...

Search for all instances of the word “some.”

tgrep2 "some"

Return all full sentences with the word “some.”

tgrep2 -w "some"

Return only the terminal nodes (i.e., the words), without parses.

tgrep2 -wt "some"

Pipe the output into less so we can scroll up and down.

tgrep2 -wt "some" | less

How many matches are there?

tgrep2 -wt "some" | wc -l

Make sure to get all subtrees matching one or more patterns, but only report each subtree once (use -af by default, it will make the results more accurate).

tgrep2 -afwt "some" | wc -l

Once we’re happy with our results, we can save them to a file.

tgrep2 -afwt "some" > "some.txt"

By default, tgrep2 searches the Switchboard (or whatever the TGREP2_CORPUS environment variable is set to). Let’s search a different corpus instead. To see the available corpora:

ls $TGREP2ABLE

Let’s search the BNC.

tgrep2 -afwtc /afs/ir/data/linguistic-data/Treebank/tgrep2able/bnc-charniak-parses.t2c.gz "some" | wc -l

Let’s go back to the Switchboard and find all the “some”-NPs. Start by looking at the parses with the -l option.

tgrep2 -aflw "some" | less

Find the NP that dominates “some.”

tgrep2 -aflw "some >> NP" | less

Oops, this still gets us just the word “some.” TGrep2 returns the first node by default. Let’s turn the pattern around.

tgrep2 -afl "NP << some" | less

The problem is that this finds any NP with “some” in it. Let’s get the closest one.

tgrep2 -afl "NP @<< NP << some" | less

To find the cases of partitive “some.”

tgrep2 -afl "NP << some << (of , some)" | less

To find the cases of non-partitive “some.”

tgrep2 -afl "NP << (some @. of)" | less

To find the cases of “some”-NPs at the start of sentences.

tgrep2 -afl "NP << some @, *" | less

To specify that the NP must contain a head noun.

tgrep2 -afl "NP=np << (some .. (/NN|NNS/ >> =np))" | less

Thus far, we have only output the results of a single search on the command line, or saved the output of a single search to a file. Ideally, we would add different types of information about each instance of “some” to a database directly – e.g., the full sentence, whether “some” occurs in the partitive, and whether “some” occurs at the start of a sentence. To do this easily, use TDTlite.