reckart
commented
9 years ago Hello everyone,
Stanford NLP is pleased to announce version 3.5.0 of our software tools.
There are two big new things in this release:
1. Version 3.5.0 includes the first release of a new high performance dependency parser.
You can use it in Stanford CoreNLP as the depparser annotator, or run it standalone
using the Stanford Parser. This is a neural network-based shift-reduce (or transition-based)
dependency parser. It can provide the same kind of fast dependency parsing as parsers
such as MaltParser, DeSR or zpar; indeed the efficient neural network model allows
it to be faster than MaltParser. It is described in the paper:
Danqi Chen and Christopher D Manning. A Fast and Accurate Dependency Parser using Neural
Networks. EMNLP 2014.
At present, this parser only supports English Stanford Dependencies. (In the parser
download, we also include models trained for CoNLL 2008 English and CoNLL 2006 Chinese.)
Our intention is to support more languages in the next release. It should also be noted
that this parser does not produce phrase structure constituents, and so at present
you cannot use it with the coreference system, which requires constituency parsers.
However, if the main thing you want is English Stanford Dependencies output, you should
definitely check it out! It’s fast.
2. Version 3.5.0 moves the Stanford NLP tools to Java 8, including making some incompatible
changes. We realize that stability and compatibility is important to people, and hence
for many years we have kept our Stanford NLP tools so that they were compiled with
and ran under Java 6+. However, we felt that we couldn’t stay on Java 6 forever, and
now seemed a good time to take advantage of the big benefits that come from the new
functional programming support in Java 8. Hence, barring finding any terrible bugs,
the end of August release, version 3.4.1 will be the final release that runs under
Java 6. This release requires Java 8.
If we were moving to Java 8, it seemed useful to then make some incompatible changes
so that our tools took advantage of Java 8. In particular, from our util classes, we
removed our Function and Filter interfaces, and replaced them with the java.util.function
Function and Predicate interfaces. This lets our code work much more smoothly with
the innovations of Java 8. And at the same time, we improved the API of our util class
Index, which was quite poor. We haven’t gone out of our way to make things incompatible,
so most code operating with the higher-level interfaces of Stanford NLP software will
work just as before, but if you are working with our code at a lower level, then you
might have to make some code changes to update your software for these changes.
Version 3.5.0 also includes a new annotator for relation extraction and improved code
for pattern learning. You can find usage information for the relation extractor here.
As usual, there are also other bug fixes and minor improvements. These include adding
the use of classifiers trained with the Stanford Classifier to CoreNLP pipelines, and
new output formats that provide CoNLL-style and JSON output from CoreNLP.
CoreNLP version 3.5.0 is available for download from the main CoreNLP page.
We will be releasing an updated Maven artifact shortly.
Enjoy!
Jon Gauthier
Stanford Natural Language Processing GroupOriginal issue reported on code.google.com by richard.eckart on 2014-10-30 06:30:12
Original issue reported on code.google.com by
richard.eckarton 2014-10-29 11:31:20