Stanford NER tagger generates 'file not found' exception with provided models

Question

I downloaded stanford NER 3.4.1, unpacked it, and tried to run named entity recognition on a local file using the default (provided) trained model. I got this:

 `java.io.FileNotFoundException: /u/nlp/data/pos_tags_are_useless/egw4-reut.512.clusters (No such file or directory) at edu.stanford.nlp.io.IOUtils.inputStreamFromFile(IOUtils.java:481)`

What's wrong and how can I fix it?

Answer 1

It turns out that the provided models use "distributional similarity features" that require a .clusters file at a location specified in the compressed model file (tricky to change). If you're on the stanford network, presumably the required files are there. If not, I found two choices:

1. Download stanford NER without the distributional similarity features (slightly degrades performance, but runs faster). disclaimer: I havn't actually tried this, but it should work.
2. Download the distsim file (look here) from stanford and create a sym-link to it so it appears to be in the correct location. In my case on a Mac, I did this:
   • I created a heirarchy of folders u/nlp/data/pos_tags_are_useless/ somewhere,
   • copied the downloaded egw4-reut.512.clusters file there,
   • then cd /; sudo ln -s <somewhere>/u.

Answer 2

This was an error in the model files accompanying the initial release of the v3.4.1 code, and has been fixed. Re-download and all should run fine, without requiring the symlink workaround.