Framenet annotation

This is a browser-based tool for annotating sentences with Framenet 1.5 frames and arguments. It starts a web server, which runs on the local machine and is only accessible there.

Screenshot shows the annotation of a tweet from the Ritter and Clark corpus

As input the tool accepts a folder of files, each containing one sentence in tab-separated format with one token per line. The final column is a space-separated list of frames that could be triggered by that token. The output looks like the input, except for the last column, which has been replaced by the annotations selected in the interface. Saving happens automatically in the background.

Here is an example of the input format (see data/demo/ritter.dev01):

But     CONJ
in      ADP
any     DET
case    NOUN    Instance Reasoning Containers Trial
I       PRON
suppose VERB    Opinion
you     PRON
will    VERB    Giving Desiring
not     ADV
let     VERB    Grant_permission Make_possible_to_do
it      PRON
away    ADV
for     ADP
some    DET
days    NOUN    Calendric_unit Timespan Measure_duration
?       .

In the standard input format, the first column has the token form and the second column has the part of speech. Additionally, the tool supports the CONLL9 format, where the token is in the second column and the part of speech is in the fourth column. CONNL9 will be automatically selected if the number of input columns is equal to 14.

Installation

The annotation tool is written in Python and depends on the web framework flask as well as a fairly recent version of NLTK.

Install the dependencies via pip:

pip install -r requirements.txt

The tool uses the NLTK distribution of the Framenet data. It looks for the data in $HOME/nltk_data/corpora/framenet_v15, which is the default install location used by NLTK. If the data is not found, an attempt to download it will be made using NLTK. Unfortunately, this will fail if your NLTK data is in a non-standard location.

Usage

To run the demo, type % python src/annotate.py data/demo .. This puts the annotated files in the current directory.

To access the web interface go to http://127.0.0.1:5000/annotate/ritter.dev01.

The command line options are:

usage: annotate.py [-h] in_dir out_dir

A web interface for annotating Framenet across languages

positional arguments:
  in_dir      Directory with input files, each containing a single sentence in
              tab separated format.The last column contains a list of space-
              separated list of possible frames evoked by that token
  out_dir     Directory for finished annotations

optional arguments:
  -h, --help  show this help message and exit

Caveat

In the current version, visiting a sentence in the browser that has already been annotated will overwrite the existing annotation.