spaCy is a library for advanced natural language processing in Python and Cython. spaCy is built on the very latest research, but it isn't researchware. It was designed from day 1 to be used in real products. It's commercial open-source software, released under the MIT license.

💫 Version 1.2 out now! Read the release notes here.

• Non-destructive tokenization
• Syntax-driven sentence segmentation
• Pre-trained word vectors
• Part-of-speech tagging
• Named entity recognition
• Labelled dependency parsing
• Convenient string-to-int mapping
• Export to numpy data arrays
• GIL-free multi-threading
• Efficient binary serialization
• Easy deep learning integration
• Statistical models for English and German
• State-of-the-art speed
• Robust, rigorously evaluated accuracy

See facts, figures and benchmarks.

• Fastest in the world: <50ms per document. No faster system has ever been announced.
• Accuracy within 1% of the current state of the art on all tasks performed (parsing, named entity recognition, part-of-speech tagging). The only more accurate systems are an order of magnitude slower or more.

• CPython 2.6, 2.7, 3.3, 3.4, 3.5 (only 64 bit)
• macOS / OS X
• Linux
• Windows (Cygwin, MinGW, Visual Studio)

spaCy is compatible with 64-bit CPython 2.6+/3.3+ and runs on Unix/Linux, OS X and Windows. Source packages are available via pip. Please make sure that you have a working build enviroment set up. See notes on Ubuntu, macOS/OS X and Windows for details.

When using pip it is generally recommended to install packages in a virtualenv to avoid modifying system state:

pip install spacy

Python packaging is awkward at the best of times, and it's particularly tricky with C extensions, built via Cython, requiring large data files. So, please report issues as you encounter them.

After installation you need to download a language model. Currently only models for English and German, named en and de, are available.

python -m spacy.en.download all
python -m spacy.de.download all

The download command fetches about 1 GB of data which it installs within the spacy package directory.

To upgrade spaCy to the latest release:

pip install -U spacy

Sometimes new releases require a new language model. Then you will have to upgrade to a new model, too. You can also force re-downloading and installing a new language model:

python -m spacy.en.download --force

The other way to install spaCy is to clone its GitHub repository and build it from source. That is the common way if you want to make changes to the code base.

You'll need to make sure that you have a development enviroment consisting of a Python distribution including header files, a compiler, pip, virtualenv and git installed. The compiler part is the trickiest. How to do that depends on your system. See notes on Ubuntu, OS X and Windows for details.

# make sure you are using recent pip/virtualenv versions
python -m pip install -U pip virtualenv

#  find git install instructions at https://git-scm.com/downloads
git clone https://github.com/explosion/spaCy.git

cd spaCy
virtualenv .env && source .env/bin/activate
pip install -r requirements.txt
pip install -e .

Compared to regular install via pip requirements.txt additionally installs developer dependencies such as cython.

Install system-level dependencies via apt-get:

sudo apt-get install build-essential python-dev git

Install a recent version of XCode, including the so-called "Command Line Tools". macOS and OS X ship with Python and git preinstalled.

Install a version of Visual Studio Express or higher that matches the version that was used to compile your Python interpreter. For official distributions these are VS 2008 (Python 2.7), VS 2010 (Python 3.4) and VS 2015 (Python 3.5).

spaCy comes with an extensive test suite. First, find out where spaCy is installed:

python -c "import os; import spacy; print(os.path.dirname(spacy.__file__))"

Then run pytest on that directory. The flags --vectors, --slow and --model are optional and enable additional tests:

# make sure you are using recent pytest version
python -m pip install -U pytest

python -m pytest <spacy-directory> --vectors --model --slow

✨ Major features and improvements

• NEW: Support Chinese tokenization, via Jieba.
• NEW: Alpha support for French, Spanish, Italian and Portuguese tokenization.

🔴 Bug fixes

• Fix issue #376: POS tags for "and/or" are now correct.
• Fix issue #578: --force argument on download command now operates correctly.
• Fix issue #595: Lemmatization corrected for some base forms.
• Fix issue #588: Matcher now rejects empty patterns.
• Fix issue #592: Added exception rule for tokenization of "Ph.D."
• Fix issue #599: Empty documents now considered tagged and parsed.
• Fix issue #600: Add missing token.tag and token.tag_ setters.
• Fix issue #596: Added missing unicode import when compiling regexes that led to incorrect tokenization.
• Fix issue #587: Resolved bug that caused Matcher to sometimes segfault.
• Fix issue #429: Ensure missing entity types are added to the entity recognizer.

• Rename new pipeline keyword argument of spacy.load() to create_pipeline.
• Rename new vectors keyword argument of spacy.load() to add_vectors.

🔴 Bug fixes

• Fix issue #544: Add vocab.resize_vectors() method, to support changing to vectors of different dimensionality.
• Fix issue #536: Default probability was incorrect for OOV words.
• Fix issue #539: Unspecified encoding when opening some JSON files.
• Fix issue #541: GloVe vectors were being loaded incorrectly.
• Fix issue #522: Similarities and vector norms were calculated incorrectly.
• Fix issue #461: ent_iob attribute was incorrect after setting entities via doc.ents
• Fix issue #459: Deserialiser failed on empty doc
• Fix issue #514: Serialization failed after adding a new entity label.

✨ Major features and improvements

• NEW: custom processing pipelines, to support deep learning workflows
• NEW: Rule matcher now supports entity IDs and attributes
• NEW: Official/documented training APIs and GoldParse class
• Download and use GloVe vectors by default
• Make it easier to load and unload word vectors
• Improved rule matching functionality
• Move basic data into the code, rather than the json files. This makes it simpler to use the tokenizer without the models installed, and makes adding new languages much easier.
• Replace file-system strings with Path objects. You can now load resources over your network, or do similar trickery, by passing any object that supports the Path protocol.

⚠️ Backwards incompatibilities

• The data_dir keyword argument of Language.__init__ (and its subclasses English.__init__ and German.__init__) has been renamed to path.
• Details of how the Language base-class and its sub-classes are loaded, and how defaults are accessed, have been heavily changed. If you have your own subclasses, you should review the changes.
• The deprecated token.repvec name has been removed.
• The .train() method of Tagger and Parser has been renamed to .update()
• The previously undocumented GoldParse class has a new __init__() method. The old method has been preserved in GoldParse.from_annot_tuples().
• Previously undocumented details of the Parser class have changed.
• The previously undocumented get_package and get_package_by_name helper functions have been moved into a new module, spacy.deprecated, in case you still need them while you update.

🔴 Bug fixes

• Fix get_lang_class bug when GloVe vectors are used.
• Fix Issue #411: doc.sents raised IndexError on empty string.
• Fix Issue #455: Correct lemmatization logic
• Fix Issue #371: Make Lexeme objects hashable
• Fix Issue #469: Make noun_chunks detect root NPs

👥 Contributors

Thanks to @daylen, @RahulKulhari, @stared, @adamhadani, @izeye and @crawfordcomeaux for the pull requests!

• Fixed bug that prevented German parses from being deprojectivised.
• Bug fixes to sentence boundary detection.
• Add rich comparison methods to the Lexeme class.
• Add missing Doc.has_vector and Span.has_vector properties.
• Add missing Span.sent property.

spaCy finally supports another language, in addition to English. We're lucky to have Wolfgang Seeker on the team, and the new German model is just the beginning. Now that there are multiple languages, you should consider loading spaCy via the load() function. This function also makes it easier to load extra word vector data for English:

import spacy
en_nlp = spacy.load('en', vectors='en_glove_cc_300_1m_vectors')
de_nlp = spacy.load('de')

To support use of the load function, there are also two new helper functions: spacy.get_lang_class and spacy.set_lang_class. Once the German model is loaded, you can use it just like the English model:

doc = nlp(u'''Wikipedia ist ein Projekt zum Aufbau einer Enzyklopädie aus freien Inhalten, zu dem du mit deinem Wissen beitragen kannst. Seit Mai 2001 sind 1.936.257 Artikel in deutscher Sprache entstanden.''')

for sent in doc.sents:
    print(sent.root.text, sent.root.n_lefts, sent.root.n_rights)

# (u'ist', 1, 2)
# (u'sind', 1, 3)

The German model provides tokenization, POS tagging, sentence boundary detection, syntactic dependency parsing, recognition of organisation, location and person entities, and word vector representations trained on a mix of open subtitles and Wikipedia data. It doesn't yet provide lemmatisation or morphological analysis, and it doesn't yet recognise numeric entities such as numbers and dates.

Bugfixes

• spaCy < 0.100.7 had a bug in the semantics of the Token.__str__ and Token.__unicode__ built-ins: they included a trailing space.
• Improve handling of "infixed" hyphens. Previously the tokenizer struggled with multiple hyphens, such as "well-to-do".
• Improve handling of periods after mixed-case tokens
• Improve lemmatization for English special-case tokens
• Fix bug that allowed spaces to be treated as heads in the syntactic parse
• Fix bug that led to inconsistent sentence boundaries before and after serialisation.
• Fix bug from deserialising untagged documents.

This release offers improved support for replacing the word vectors used by spaCy. To install Stanford's GloVe vectors, trained on the Common Crawl, just run:

sputnik --name spacy install en_glove_cc_300_1m_vectors

To reduce memory usage and loading time, we've trimmed the vocabulary down to 1m entries.

This release also integrates all the code necessary for German parsing. A German model will be released shortly. To assist in multi-lingual processing, we've added a load() function. To load the English model with the GloVe vectors:

spacy.load('en', vectors='en_glove_cc_300_1m_vectors')

Fix incorrect use of header file, caused from problem with thinc

Small correction to right_edge calculation

Support multi-threading, via the .pipe method. spaCy now releases the GIL around the parser and entity recognizer, so systems that support OpenMP should be able to do shared memory parallelism at close to full efficiency.

We've also greatly reduced loading time, and fixed a number of bugs.

Fix data version lock that affected v0.100.1

v0.100 included header files built on Linux that caused installation to fail on OSX. This should now be corrected. We also update the default data distribution, to include a small fix to the tokenizer.

• Redo setup.py, and remove ugly headers_workaround hack. Should result in fewer install problems.
• Update data downloading and installation functionality, by migrating to the Sputnik data-package manager. This will allow us to offer finer grained control of data installation in future.
• Fix bug when using custom entity types in Matcher. This should work by default when using the English.__call__ method of running the pipeline. If invoking Parser.__call__ directly to do NER, you should call the Parser.add_label() method to register your entity type.
• Fix head-finding rules in Span.
• Fix problem that caused doc.merge() to sometimes hang
• Fix problems in handling of whitespace

• Merging multi-word tokens into one, via the doc.merge() and span.merge() methods, no longer invalidates existing Span objects. This makes it much easier to merge multiple spans, e.g. to merge all named entities, or all base noun phrases. Thanks to @andreasgrv for help on this patch.
• Lots of internal refactoring, especially around the machine learning module, thinc. The thinc API has now been improved, and the spacy._ml wrapper module is no longer necessary.
• The lemmatizer now lower-cases non-noun, noun-verb and non-adjective words.
• A new attribute, .rank, is added to Token and Lexeme objects, giving the frequency rank of the word.

• Remove binary data from PyPi package.
• Delete archive after downloading data
• Use updated cymem, preshed and thinc packages
• Fix information loss in deserialize
• Fix __str__ methods for Python2

• Fix bugs in download.py
• Require --force to over-write the data directory in download.py
• Fix bugs in Matcher and doc.merge()

• Fix bug that caused text to be lost after .merge
• Fix bug in Matcher when matched entities overlapped

• Reform encoding of symbols
• Fix bugs in Matcher
• Fix bugs in Span
• Add tokenizer rule to fix numeric range tokenization
• Add specific string-length cap in Tokenizer
• Fix token.conjuncts

• Fix memory error that caused crashes on 32bit platforms
• Fix parse errors caused by smart quotes and em-dashes

Bug fixes to word vectors