About | Citation | Getting started | Questions | Mailing lists | Download | Extensions | Models | Online demo | Release history | FAQ
Stanford NER is a Java implementation of a Named Entity Recognizer. Named Entity Recognition (NER) labels sequences of words in a text which are the names of things, such as person and company names, or gene and protein names. It comes with well-engineered feature extractors for Named Entity Recognition, and many options for defining feature extractors. Included with the download are good named entity recognizers for English, particularly for the 3 classes (PERSON, ORGANIZATION, LOCATION), and we also make available on this page various other models for different languages and circumstances, including models trained on just the CoNLL 2003 English training data.
Stanford NER is also known as CRFClassifier. The software provides a general implementation of (arbitrary order) linear chain Conditional Random Field (CRF) sequence models. That is, by training your own models on labeled data, you can actually use this code to build sequence models for NER or any other task. (CRF models were pioneered by Lafferty, McCallum, and Pereira (2001); see Sutton and McCallum (2006) or Sutton and McCallum (2010) for more comprehensible introductions.)
The original CRF code is by Jenny Finkel. The feature extractors are by Dan Klein, Christopher Manning, and Jenny Finkel. Much of the documentation and usability is due to Anna Rafferty. More recent code development has been done by various Stanford NLP Group members.
Stanford NER is available for download,
licensed under the GNU
General Public License (v2 or later). Source is included.
The package includes components for command-line invocation (look at the
shell scripts and batch files included in the download), running as a
server (look at NERServer in the sources jar file), and a
Java API (look at the simple examples in the
NERDemo.java file
included in the download, and then at the javadocs).
Stanford NER
code is dual licensed (in a similar manner to MySQL, etc.).
Open source licensing is under the full GPL,
which allows many free uses.
For distributors of
proprietary
software, commercial licensing is available.
If you don't need a commercial license, but would like to support
maintenance of these tools, we welcome gifts.
The CRF sequence models provided here do not precisely correspond to any published paper, but the correct paper to cite for the model and software is:
Jenny Rose Finkel, Trond Grenager, and Christopher
Manning. 2005. Incorporating Non-local Information into Information
Extraction Systems by Gibbs Sampling.
Proceedings of the 43nd Annual Meeting of the Association for Computational Linguistics (ACL 2005),
pp. 363-370.
http://nlp.stanford.edu/~manning/papers/gibbscrf3.pdf
The software provided here is similar to the baseline local+Viterbi
model in that paper, but adds new
distributional similarity based features (in the -distSim
classifiers).
Distributional similarity features improve performance but the models
require somewhat more memory.
Our big English NER models were trained on a mixture of CoNLL, MUC-6, MUC-7
and ACE named entity corpora, and as a result the models are fairly robust
across domains.
You can try out Stanford NER CRF classifiers or Stanford NER as part of Stanford CoreNLP on the web, to understand what Stanford NER is and whether it will be useful to you.
To use the software on your computer, download the zip file.
You then unzip the file by either double-clicing on the zip file, using a program for unpacking zip files, or by using
the unzip command. This shord create a stanford-ner folder. There is no installation procedure, you should be able to run Stanford NER from that folder. Normally, Stanford NER is run from the command line (i.e., shell or terminal).
Current releases of Stanford NER require Java 1.8 or later. Either make sure you have or get Java 8
or consider running an earlier version of the software (versions through 3.4.1 support Java 6 and 7)..
From a command line, you need to have java on your PATH and the
stanford-ner.jar file in your CLASSPATH. (The way of doing this depends on
your OS/shell.) The supplied ner.bat and ner.sh should work to allow
you to tag a single file, when running from inside the Stanford NER folder. For example, for Windows:
ner file
This corresponds to the full command:
java -mx600m -cp "*;lib\*" edu.stanford.nlp.ie.crf.CRFClassifier -loadClassifier classifiers/english.all.3class.distsim.crf.ser.gz -textFile sample.txt
Or on Unix/Linux you should be able to parse the test file in the distribution directory with the command:
java -mx600m -cp "*:lib/*" edu.stanford.nlp.ie.crf.CRFClassifier -loadClassifier classifiers/english.all.3class.distsim.crf.ser.gz -textFile sample.txt
Here's an output option that will print out entities and their class to the first two columns of a tab-separated columns output file:
java -mx600m -cp "*;lib/*" edu.stanford.nlp.ie.crf.CRFClassifier -loadClassifier classifiers/english.all.3class.distsim.crf.ser.gz -outputFormat tabbedEntities -textFile sample.txt > sample.tsv
This standalone distribution also allows access to the full NER
capabilities of the Stanford CoreNLP pipeline. These capabilities
can be accessed via the NERClassifierCombiner class.
NERClassifierCombiner allows for multiple CRFs to be used together,
and has options for recognizing numeric sequence patterns and time
patterns with the rule-based NER of SUTime.
To use NERClassifierCombiner at the command-line, the jars in lib directory and stanford-ner.jar must be in the CLASSPATH. Here is an example command:
java -mx1g -cp "*:lib/*" edu.stanford.nlp.ie.NERClassifierCombiner -textFile
sample.txt -ner.model
classifiers/english.all.3class.distsim.crf.ser.gz,classifiers/english.conll.4class.distsim.crf.ser.gz,classifiers/english.muc.7class.distsim.crf.ser.gz
The one difference you should see from above is that Sunday is now recognized as a DATE.
You can call Stanford NER from your own code. The
file NERDemo.java included in the distribution illustrates
several ways of calling the system programatically.
We suggest that you start from there, and then look at the javado,
etc. as needed.
Stanford NER can also be set up to run as a server listening on a socket.
You can look at a Powerpoint Introduction to NER and the Stanford NER
package [ppt]
[pdf].
There is also a list of Frequently Asked
Questions (FAQ), with answers! This includes
some information on training models.
Further documentation is provided in the included
README.txt and in the javadocs.
Have a support question? Ask us on Stack Overflow using the tag stanford-nlp.
Feedback and bug reports / fixes can be sent to our mailing lists.
We have 3 mailing lists for the Stanford Named Entity Recognizer,
all of which are shared
with other JavaNLP tools (with the exclusion of the parser). Each address is
at @lists.stanford.edu:
java-nlp-user This is the best list to post to in order
to send feature requests, make announcements, or for discussion among JavaNLP
users. (Please ask support questions on
Stack Overflow using the
stanford-nlp tag.)
You have to subscribe to be able to use this list.
Join the list via this webpage or by emailing
java-nlp-user-join@lists.stanford.edu. (Leave the
subject and message body empty.) You can also
look at
the list archives.
java-nlp-announce This list will be used only to announce
new versions of Stanford JavaNLP tools. So it will be very low volume (expect 1-3
messages a year). Join the list via this webpage or by emailing
java-nlp-announce-join@lists.stanford.edu. (Leave the
subject and message body empty.)
java-nlp-support This list goes only to the software
maintainers. It's a good address for licensing questions, etc. For
general use and support questions, you're better off joining and using
java-nlp-user.
You cannot join java-nlp-support, but you can mail questions to
java-nlp-support@lists.stanford.edu.
The download is a 151M zipped file (mainly consisting of classifier data objects). If you unpack that file, you should have everything needed for English NER (or use as a general CRF). It includes batch files for running under Windows or Unix/Linux/MacOSX, a simple GUI, and the ability to run as a server. Stanford NER requires Java v1.8+. If you want to use Stanford NER for other languages, you'll also need to download model files for those languages; see further below.
For some (computer) languages, there are more up-to-date interfaces to Stanford NER available by using it inside Stanford CoreNLP, and you are better off getting those from the CoreNLP page and using them....
Included with Stanford NER are a 4 class model trained on the CoNLL 2003 eng.train, a 7 class model trained on the MUC 6 and MUC 7 training data sets, and a 3 class model trained on both data sets and some additional data (including ACE 2002 and limited amounts of in-house data) on the intersection of those class sets. (The training data for the 3 class model does not include any material from the CoNLL eng.testa or eng.testb data sets, nor any of the MUC 6 or 7 test or devtest datasets, nor Alan Ritter's Twitter NER data, so all of these remain valid tests of its performance.)
| 3 class: | Location, Person, Organization |
| 4 class: | Location, Person, Organization, Misc |
| 7 class: | Location, Person, Organization, Money, Percent, Date, Time |
These models each use distributional similarity features, which
provide considerable performance gain at the cost of increasing their size and
runtime. We also have models that are the same except without the distributional similarity features.
You can find them in our English models jar. You can
either unpack the jar file or add it to the classpath; if you add the
jar file to the classpath, you can then load the models from the path
edu/stanford/nlp/models/.... You can run
jar -tf <jar-file> to get the list of files in the jar file.
Also available are caseless versions of these models, better for use on texts that are mainly lower or upper case, rather than follow the conventions of standard English
Important note: There was a problem with the v3.6.0 English Caseless NER model. See this page.
A German NER model is available, based on work by Manaal Faruqui and Sebastian Padó. You can find it in the CoreNLP German models jar. For citation and other information relating to the German classifiers, please see Sebastian Pado's German NER page (but the models there are now many years old; you should use the better models that we have!). It is a 4 class IOB1 classifier (see, e.g., Memory-Based Shallow Parsing by Erik F. Tjong Kim Sang). The tags given to words are: I-LOC, I-PER, I-ORG, I-MISC, B-LOC, B-PER, B-ORG, B-MISC, O. It is trained over the CoNLL 2003 data with distributional similarity classes built from the Huge German Corpus.
German resources (written in German): a help page for Stanford NER; a workshop on using Stanford CoreNLP for NER with materials available.
Here are a couple of commands using these models, two sample files, and a couple of
notes. Running on TSV files: the models were saved with options for testing on German CoNLL NER
files. While the models use just the surface word form, the input reader
expects the word in the first column and the class in the fifth colum
(1-indexed colums). You can either make the input like that or else
change the expectations with, say, the option -map "word=0,answer=1" (0-indexed columns).
These models were also trained on data with straight ASCII quotes and
BIO entity tags. Also, be careful of the text encoding: The default is
Unicode; use -encoding iso-8859-15 if the text is in 8-bit encoding.
TSV mini test file:german-ner.tsv— Text mini test file:german-ner.txt
java -cp "*" edu.stanford.nlp.ie.crf.CRFClassifier -loadClassifier edu/stanford/nlp/models/ner/german.conll.hgc_175m_600.crf.ser.gz -testFile german-ner.tsv java -cp "*" edu.stanford.nlp.ie.crf.CRFClassifier -loadClassifier edu/stanford/nlp/models/ner/german.conll.hgc_175m_600.crf.ser.gz -tokenizerOptions latexQuotes=false -textFile german-ner.txt
From version 3.4.1 forward, we have a Spanish model available for NER. It is included in the Spanish corenlp models jar.
We also provide Chinese models built from the Ontonotes Chinese named entity data. There are two models, one using distributional similarity clusters and one without. These are designed to be run on word-segmented Chinese. So, if you want to use these on normal Chinese text, you will first need to run Stanford Word Segmenter or some other Chinese word segmenter, and then run NER on the output of that!
We have an online demo of several of our NER models. Special thanks to Dat Hoang, who provided the initial version. Note that the online demo demonstrates single CRF models; in order to see the effect of the time annotator or the combined models, see CoreNLP.
| Version | Date | Description |
|---|---|---|
| 4.2.0 | 2020-11-17 | Update for compatibility |
| 4.0.0 | 2020-04-19 | Update to UDv2.0 tokenization |
| 3.9.2 | 2018-10-16 | Updated for compatibility |
| 3.9.1 | 2018-02-27 | KBP ner models for Chinese and Spanish |
| 3.8.0 | 2017-06-09 | Updated for compatibility |
| 3.7.0 | 2016-10-31 | Improvements to Chinese and German NER |
| 3.6.0 | 2015-12-09 | Updated for compatibility |
| 3.5.2 | 2015-04-20 | synch standalone and CoreNLP functionality |
| 3.5.1 | 2015-01-29 | Substantial accuracy improvements |
| 3.5.0 | 2014-10-26 | Upgrade to Java 8 |
| 3.4.1 | Added Spanish models | |
| 3.4 | Fix serialization of new models | |
| 3.3.1 | Bugfix release | |
| 3.3.0 | Updated for compatibility | |
| 3.2.0 | Improved line by line handling | |
| 1.2.8 | -nthreads option | |
| 1.2.7 | Add Chinese model, include Wikipedia data in 3-class English model | |
| 1.2.6 | Minor bug fixes | |
| 1.2.5 | Fix encoding issue | |
| 1.2.4 | Caseless versions of models supported | |
| 1.2.3 | Minor bug fixes | |
| 1.2.2 | Improved thread safety | |
| 1.2.1 | Models reduced in size but on average improved in accuracy (improved distsim clusters) | |
| 1.2 | Normal download includes 3, 4, and 7 class models. Updated for compatibility with other software releases. | |
| 1.1.1 | Minor bug and usability fixes, and changed API (in particular the methods to classify and output tagged text) | |
| 1.1 | Additional feature flags, various code updates | |
| 1.0 | Initial release |