spaCy/website/docs/usage/adding-languages.jade

//- 💫 DOCS > USAGE > ADDING LANGUAGES

include ../../_includes/_mixins

p
        |  Adding full support for a language touches many different parts of the
        |  spaCy library. This guide explains how to fit everything together, and
        |  points you to the specific workflows for each component.

+aside("Working on spaCy's source")
    |  To add a new language to spaCy, you'll need to
    |  #[strong modify the library's code]. The easiest way to do this is to
    |  clone the #[+src(gh("spaCy")) repository] and #[strong build spaCy from source].
    |  For more information on this, see the #[+a("/docs/usage") installation guide].
    |  Unlike spaCy's core, which is mostly written in Cython, all language
    |  data is stored in regular Python files. This means that you won't have to
    |  rebuild anything in between – you can simply make edits and reload spaCy
    |  to test them.

+grid.o-no-block
    +grid-col("half")
        p
            |  Obviously, there are lots of ways you can organise your code when
            |  you implement your own language data. This guide will focus on
            |  how it's done within spaCy. For full language support, you'll
            |  need to create a #[code Language] subclass, define custom
            |  #[strong language data], like a stop list and tokenizer
            |  exceptions and test the new tokenizer. Once the language is set
            |  up, you can #[strong build the vocabulary], including word
            |  frequencies, Brown clusters and word vectors. Finally, you can
            |  #[strong train the tagger and parser], and save the model to a
            |  directory.

        p
            |  For some languages, you may also want to develop a solution for
            |  lemmatization and morphological analysis.

    +table-of-contents
        +item #[+a("#101") Language data 101]
        +item #[+a("#language-subclass") The Language subclass]
        +item #[+a("#stop-words") Stop words]
        +item #[+a("#tokenizer-exceptions") Tokenizer exceptions]
        +item #[+a("#norm-exceptions") Norm exceptions]
        +item #[+a("#lex-attrs") Lexical attributes]
        +item #[+a("#lemmatizer") Lemmatizer]
        +item #[+a("#tag-map") Tag map]
        +item #[+a("#morph-rules") Morph rules]
        +item #[+a("#testing") Testing the tokenizer]
        +item #[+a("#vocabulary") Building the vocabulary]
        +item #[+a("#training") Training]

+h(2, "101") Language data 101

include _spacy-101/_language-data

p
    |  The individual components #[strong expose variables] that can be imported
    |  within a language module, and added to the language's #[code Defaults].
    |  Some components, like the punctuation rules, usually don't need much
    |  customisation and can simply be imported from the global rules. Others,
    |  like the tokenizer and norm exceptions, are very specific and will make
    |  a big difference to spaCy's performance on the particular language and
    |  training a language model.


+table(["Variable", "Type", "Description"])
    +row
        +cell #[code STOP_WORDS]
        +cell set
        +cell Individual words.

    +row
        +cell #[code TOKENIZER_EXCEPTIONS]
        +cell dict
        +cell Keyed by strings mapped to list of one dict per token with token attributes.

    +row
        +cell #[code TOKEN_MATCH]
        +cell regex
        +cell Regexes to match complex tokens, e.g. URLs.

    +row
        +cell #[code NORM_EXCEPTIONS]
        +cell dict
        +cell Keyed by strings, mapped to their norms.

    +row
        +cell #[code TOKENIZER_PREFIXES]
        +cell list
        +cell Strings or regexes, usually not customised.

    +row
        +cell #[code TOKENIZER_SUFFIXES]
        +cell list
        +cell Strings or regexes, usually not customised.

    +row
        +cell #[code TOKENIZER_INFIXES]
        +cell list
        +cell Strings or regexes, usually not customised.

    +row
        +cell #[code LEX_ATTRS]
        +cell dict
        +cell Attribute ID mapped to function.

    +row
        +cell #[code LOOKUP]
        +cell dict
        +cell Keyed by strings mapping to their lemma.

    +row
        +cell #[code LEMMA_RULES], #[code LEMMA_INDEX], #[code LEMMA_EXC]
        +cell dict
        +cell Lemmatization rules, keyed by part of speech.

    +row
        +cell #[code TAG_MAP]
        +cell dict
        +cell
            |  Keyed by strings mapped to
            |  #[+a("http://universaldependencies.org/u/pos/all.html") Universal Dependencies]
            |  tags.

    +row
        +cell #[code MORPH_RULES]
        +cell dict
        +cell Keyed by strings mapped to a dict of their morphological features.

+aside("Should I ever update the global data?")
    |  Reuseable language data is collected as atomic pieces in the root of the
    |  #[+src(gh("spaCy", "lang")) spacy.lang] package. Often, when a new
    |  language is added, you'll find a pattern or symbol that's missing. Even
    |  if it isn't common in other languages, it might be best to add it to the
    |  shared language data, unless it has some conflicting interpretation. For
    |  instance, we don't expect to see guillemot quotation symbols
    |  (#[code »] and #[code «]) in English text. But if we do see
    |  them, we'd probably prefer the tokenizer to split them off.

+infobox("For languages with non-latin characters")
    |  In order for the tokenizer to split suffixes, prefixes and infixes, spaCy
    |  needs to know the language's character set. If the language you're adding
    |  uses non-latin characters, you might need to add the required character
    |  classes to the global
    |  #[+src(gh("spacy", "spacy/lang/char_classes.py")) char_classes.py].
    |  spaCy uses the #[+a("https://pypi.python.org/pypi/regex/") #[code regex] library]
    |  to keep this simple and readable. If the language requires very specific
    |  punctuation rules, you should consider overwriting the default regular
    |  expressions with your own in the language's #[code Defaults].


+h(2, "language-subclass") Creating a #[code Language] subclass

p
    |  Language-specific code and resources should be organised into a
    |  subpackage of spaCy, named according to the language's
    |  #[+a("https://en.wikipedia.org/wiki/List_of_ISO_639-1_codes") ISO code].
    |  For instance, code and resources specific to Spanish are placed into a
    |  directory #[code spacy/lang/es], which can be imported as
    |  #[code spacy.lang.es].

p
    |  To get started, you can use our
    |  #[+src(gh("spacy-dev-resources", "templates/new_language")) templates]
    |  for the most important files. Here's what the class template looks like:

+code("__init__.py (excerpt)").
    # import language-specific data
    from .stop_words import STOP_WORDS
    from .tokenizer_exceptions import TOKENIZER_EXCEPTIONS
    from .lex_attrs import LEX_ATTRS

    from ..tokenizer_exceptions import BASE_EXCEPTIONS
    from ...language import Language
    from ...attrs import LANG
    from ...util import update_exc

    # create Defaults class in the module scope (necessary for pickling!)
    class XxxxxDefaults(Language.Defaults):
        lex_attr_getters = dict(Language.Defaults.lex_attr_getters)
        lex_attr_getters[LANG] = lambda text: 'xx' # language ISO code

        # optional: replace flags with custom functions, e.g. like_num()
        lex_attr_getters.update(LEX_ATTRS)

        # merge base exceptions and custom tokenizer exceptions
        tokenizer_exceptions = update_exc(BASE_EXCEPTIONS, TOKENIZER_EXCEPTIONS)
        stop_words = set(STOP_WORDS)

    # create actual Language class
    class Xxxxx(Language):
        lang = 'xx' # language ISO code
        Defaults = XxxxxDefaults # override defaults

    # set default export – this allows the language class to be lazy-loaded
    __all__ = ['Xxxxx']

+infobox("Why lazy-loading?")
    |  Some languages contain large volumes of custom data, like lemmatizer
    |  loopup tables, or complex regular expression that are expensive to
    |  compute. As of spaCy v2.0, #[code Language] classes are not imported on
    |  initialisation and are only loaded when you import them directly, or load
    |  a model that requires a language to be loaded. To lazy-load languages in
    |  your application, you can use the
    |  #[+api("util#get_lang_class") #[code util.get_lang_class()]] helper
    |  function with the two-letter language code as its argument.

+h(3, "stop-words") Stop words

p
    |  A #[+a("https://en.wikipedia.org/wiki/Stop_words") "stop list"] is a
    |  classic trick from the early days of information retrieval when search
    |  was largely about keyword presence and absence. It is still sometimes
    |  useful today to filter out common words from a bag-of-words model. To
    |  improve readability, #[code STOP_WORDS] are separated by spaces and
    |  newlines, and added as a multiline string.

+aside("What does spaCy consider a stop word?")
    |  There's no particularly principled logic behind what words should be
    |  added to the stop list. Make a list that you think might be useful
    |  to people and is likely to be unsurprising. As a rule of thumb, words
    |  that are very rare are unlikely to be useful stop words.

+code("Example").
    STOP_WORDS = set("""
    a about above across after afterwards again against all almost alone along
    already also although always am among amongst amount an and another any anyhow
    anyone anything anyway anywhere are around as at

    back be became because become becomes becoming been before beforehand behind
    being below beside besides between beyond both bottom but by
    """).split())

+infobox("Important note")
    |  When adding stop words from an online source, always #[strong include the link]
    |  in a comment. Make sure to #[strong proofread] and double-check the words
    |  carefully. A lot of the lists available online have been passed around
    |  for years and often contain mistakes, like unicode errors or random words
    |  that have once been added for a specific use case, but don't actually
    |  qualify.

+h(3, "tokenizer-exceptions") Tokenizer exceptions

p
    |  spaCy's #[+a("/docs/usage/customizing-tokenizer#how-tokenizer-works") tokenization algorithm]
    |  lets you deal with whitespace-delimited chunks separately. This makes it
    |  easy to define special-case rules, without worrying about how they
    |  interact with the rest of the tokenizer. Whenever the key string is
    |  matched, the special-case rule is applied, giving the defined sequence of
    |  tokens. You can also attach attributes to the subtokens, covered by your
    |  special case, such as the subtokens #[code LEMMA] or #[code TAG].

p
    |  Tokenizer exceptions can be added in the following format:

+code("tokenizer_exceptions.py (excerpt)").
    TOKENIZER_EXCEPTIONS = {
        "don't": [
            {ORTH: "do", LEMMA: "do"},
            {ORTH: "n't", LEMMA: "not", NORM: "not", TAG: "RB"}]
    }

+infobox("Important note")
    |  If an exception consists of more than one token, the #[code ORTH] values
    |  combined always need to #[strong match the original string]. The way the
    |  original string is split up can be pretty arbitrary sometimes – for
    |  example "gonna" is split into "gon" (lemma "go") nad "na" (lemma "to").
    |  Because of how the tokenizer works, it's currently not possible to split
    |  single-letter strings into multiple tokens.

p
    |  Unambiguous abbreviations, like month names or locations in English,
    |  should be added to exceptions with a lemma assigned, for example
    |  #[code {ORTH: "Jan.", LEMMA: "January"}]. Since the exceptions are
    |  added in Python, you can use custom logic to generate them more
    |  efficiently and make your data less verbose. How you do this ultimately
    |  depends on the language. Here's an example of how exceptions for time
    |  formats like "1a.m." and "1am" are generated in the English
    |  #[+src(gh("spaCy", "spacy/en/lang/tokenizer_exceptions.py")) tokenizer_exceptions.py]:

+code("tokenizer_exceptions.py (excerpt)").
    # use short, internal variable for readability
    _exc = {}

    for h in range(1, 12 + 1):
        for period in ["a.m.", "am"]:
            # always keep an eye on string interpolation!
            _exc["%d%s" % (h, period)] = [
                {ORTH: "%d" % h},
                {ORTH: period, LEMMA: "a.m."}]
        for period in ["p.m.", "pm"]:
            _exc["%d%s" % (h, period)] = [
                {ORTH: "%d" % h},
                {ORTH: period, LEMMA: "p.m."}]

    # only declare this at the bottom
    TOKENIZER_EXCEPTIONS = dict(_exc)

+aside("Generating tokenizer exceptions")
    |  Keep in mind that generating exceptions only makes sense if there's a
    |  clearly defined and #[strong finite number] of them, like common
    |  contractions in English. This is not always the case – in Spanish for
    |  instance, infinitive or imperative reflexive verbs and pronouns are one
    |  token (e.g. "vestirme"). In cases like this, spaCy shouldn't be
    |  generating exceptions for #[em all verbs]. Instead, this will be handled
    |  at a later stage during lemmatization.

p
    |  When adding the tokenizer exceptions to the #[code Defaults], you can use
    |  the #[+api("util#update_exc") #[code update_exc()]] helper function to merge
    |  them with the global base exceptions (including one-letter abbreviations
    |  and emoticons). The function performs a basic check to make sure
    |  exceptions are provided in the correct format. It can take any number of
    |  exceptions dicts as its arguments, and will update and overwrite the
    |  exception in this order. For example, if your language's tokenizer
    |  exceptions include a custom tokenization pattern for "a.", it will
    |  overwrite the base exceptions with the language's custom one.

+code("Example").
    from ...util import update_exc

    BASE_EXCEPTIONS =  {"a.": [{ORTH: "a."}], ":)": [{ORTH: ":)"}]}
    TOKENIZER_EXCEPTIONS = {"a.": [{ORTH: "a.", LEMMA: "all"}]}

    tokenizer_exceptions = update_exc(BASE_EXCEPTIONS, TOKENIZER_EXCEPTIONS)
    # {"a.": [{ORTH: "a.", LEMMA: "all"}], ":)": [{ORTH: ":)"}]}

+infobox("About spaCy's custom pronoun lemma")
    |  Unlike verbs and common nouns, there's no clear base form of a personal
    |  pronoun. Should the lemma of "me" be "I", or should we normalize person
    |  as well, giving "it" — or maybe "he"? spaCy's solution is to introduce a
    |  novel symbol, #[code.u-nowrap -PRON-], which is used as the lemma for
    |  all personal pronouns.

+h(3, "norm-exceptions") Norm exceptions

p
    |  In addition to #[code ORTH] or #[code LEMMA], tokenizer exceptions can
    |  also set a #[code NORM] attribute. This is useful to specify a normalised
    |  version of the token – for example, the norm of "n't" is "not". By default,
    |  a token's norm equals its lowercase text. If the lowercase spelling of a
    |  word exists, norms should always be in lowercase.

+aside-code("Accessing norms").
    doc = nlp(u"I can't")
    assert [t.norm_ for t in doc] == ['i', 'can', 'not']

p
    |  spaCy usually tries to normalise words with different spellings to a single,
    |  common spelling. This has no effect on any other token attributes, or
    |  tokenization in general, but it ensures that
    |  #[strong equivalent tokens receive similar representations]. This can
    |  improve the model's predictions on words that weren't common in the
    |  training data, but are equivalent to other words – for example, "realize"
    |  and "realise", or "thx" and "thanks".

p
    |  Similarly, spaCy also includes
    |  #[+src(gh("spaCy", "spacy/lang/norm_exceptions.py")) global base norms]
    |  for normalising different styles of quotation marks and currency
    |  symbols. Even though #[code $] and #[code €] are very different, spaCy
    |  normalises them both to #[code $]. This way, they'll always be seen as
    |  similar, no matter how common they were in the training data.

p
    |  Norm exceptions can be provided as a simple dictionary. For more examples,
    |  see the English
    |  #[+src(gh("spaCy", "spacy/lang/en/norm_exceptions.py")) norm_exceptions.py].

+code("Example").
    NORM_EXCEPTIONS = {
        "cos": "because",
        "fav": "favorite",
        "accessorise": "accessorize",
        "accessorised": "accessorized"
    }

p
    |  To add the custom norm exceptions lookup table, you can use the
    |  #[code add_lookups()] helper functions. It takes the default attribute
    |  getter function as its first argument, plus a variable list of
    |  dictionaries. If a string's norm is found in one of the dictionaries,
    |  that value is used – otherwise, the default function is called and the
    |  token is assigned its default norm.

+code.
    lex_attr_getters[NORM] = add_lookups(Language.Defaults.lex_attr_getters[NORM],
                                         NORM_EXCEPTIONS, BASE_NORMS)

p
    |  The order of the dictionaries is also the lookup order – so if your
    |  language's norm exceptions overwrite any of the global exceptions, they
    |  should be added first. Also note that the tokenizer exceptions will
    |  always have priority over the atrribute getters.

+h(3, "lex-attrs") Lexical attributes

p
    |  spaCy provides a range of #[+api("token#attributes") #[code Token] attributes]
    |  that return useful information on that token – for example, whether it's
    |  uppercase or lowercase, a left or right punctuation mark, or whether it
    |  resembles a number or email address. Most of these functions, like
    |  #[code is_lower] or #[code like_url] should be language-independent.
    |  Others, like #[code like_num] (which includes both digits and number
    |  words), requires some customisation.

+aside("Best practices")
    |  Keep in mind that those functions are only intended to be  an approximation.
    |  It's always better to prioritise simplicity and performance over covering
    |  very specific edge cases.#[br]#[br]
    |  English number words are pretty simple, because even large numbers
    |  consist of individual tokens, and we can get away with splitting and
    |  matching strings against a list. In other languages, like German, "two
    |  hundred and thirty-four" is one word, and thus one token. Here, it's best
    |  to match a string against a list of number word fragments (instead of a
    |  technically almost infinite list of possible number words).

p
    |  Here's an example from the English
    |  #[+src(gh("spaCy", "spacy/en/lang/lex_attrs.py")) lex_attrs.py]:

+code("lex_attrs.py").
    _num_words = ['zero', 'one', 'two', 'three', 'four', 'five', 'six', 'seven',
                  'eight', 'nine', 'ten', 'eleven', 'twelve', 'thirteen', 'fourteen',
                  'fifteen', 'sixteen', 'seventeen', 'eighteen', 'nineteen', 'twenty',
                  'thirty', 'forty', 'fifty', 'sixty', 'seventy', 'eighty', 'ninety',
                  'hundred', 'thousand', 'million', 'billion', 'trillion', 'quadrillion',
                  'gajillion', 'bazillion']

    def like_num(text):
        text = text.replace(',', '').replace('.', '')
        if text.isdigit():
            return True
        if text.count('/') == 1:
            num, denom = text.split('/')
            if num.isdigit() and denom.isdigit():
                return True
        if text in _num_words:
            return True
        return False

    LEX_ATTRS = {
        LIKE_NUM: like_num
    }

p
    |  By updating the default lexical attributes with a custom #[code LEX_ATTRS]
    |  dictionary in the language's defaults via
    |  #[code lex_attr_getters.update(LEX_ATTRS)], only the new custom functions
    |  are overwritten.

+h(3, "lemmatizer") Lemmatizer

p
    |  As of v2.0, spaCy supports simple lookup-based lemmatization. This is
    |  usually the quickest and easiest way to get started. The data is stored
    |  in a dictionary mapping a string to its lemma. To determine a token's
    |  lemma, spaCy simply looks it up in the table. Here's an example from
    |  the Spanish language data:

+code("lang/es/lemmatizer.py (excerpt)").
    LOOKUP = {
        "aba": "abar",
        "ababa": "abar",
        "ababais": "abar",
        "ababan": "abar",
        "ababanes": "ababán",
        "ababas": "abar",
        "ababoles": "ababol",
        "ababábites": "ababábite"
    }

p
    |  To add a lookup lemmatizer to your language, import the #[code LOOKUP]
    |  table and #[code Lemmatizer], and create a new classmethod:


+code("__init__py (excerpt)").
    # other imports here, plus lookup table and lookup lemmatizer
    from .lemmatizer import LOOKUP
    from ...lemmatizerlookup import Lemmatizer

    class Xxxxx(Language):
        lang = 'xx'

        class Defaults(Language.Defaults):
            # other language defaults here

            @classmethod
            def create_lemmatizer(cls, nlp=None):
                return Lemmatizer(LOOKUP)

+h(3, "tag-map") Tag map

p
    |  Most treebanks define a custom part-of-speech tag scheme, striking a
    |  balance between level of detail and ease of prediction.  While it's
    |  useful to have custom tagging schemes, it's also useful to have a common
    |  scheme, to which the more specific tags can be related. The tagger can
    |  learn a tag scheme with any arbitrary symbols. However, you need to
    |  define how those symbols map down to the
    |  #[+a("http://universaldependencies.org/u/pos/all.html") Universal Dependencies tag set].
    |  This is done by providing a tag map.

p
    |  The keys of the tag map should be #[strong strings in your tag set]. The
    |  values should be a dictionary. The dictionary must have an entry POS
    |  whose value is one of the
    |  #[+a("http://universaldependencies.org/u/pos/all.html") Universal Dependencies]
    |  tags. Optionally, you can also include morphological features or other
    |  token attributes in the tag map as well. This allows you to do simple
    |  #[+a("/docs/usage/pos-tagging#rule-based-morphology") rule-based morphological analysis].

+code("Example").
    from ..symbols import POS, NOUN, VERB, DET

    TAG_MAP = {
        "NNS":  {POS: NOUN, "Number": "plur"},
        "VBG":  {POS: VERB, "VerbForm": "part", "Tense": "pres", "Aspect": "prog"},
        "DT":   {POS: DET}
    }

+h(3, "morph-rules") Morph rules

+under-construction

+h(2, "testing") Testing the new language tokenizer

p
    |  Before using the new language or submitting a
    |  #[+a(gh("spaCy") + "/pulls") pull request] to spaCy, you should make sure
    |  it works as expected. This is especially important if you've added custom
    |  regular expressions for token matching or punctuation – you don't want to
    |  be causing regressions.

+aside("spaCy's test suite")
    |  spaCy uses the #[+a("https://docs.pytest.org/en/latest/") pytest framework]
    |  for testing. For more details on how the tests are structured and best
    |  practices for writing your own tests, see our
    |  #[+a(gh("spaCy", "spacy/tests")) tests documentation].

+h(3, "testing-tokenizer") Testing the basic tokenizer

p
    |  The easiest way to test your new tokenizer is to run the
    |  language-independent "tokenizer sanity" tests located in
    |  #[+src(gh("spaCy", "spacy/tests/tokenizer")) tests/tokenizer]. This will
    |  test for basic behaviours like punctuation splitting, URL matching and
    |  correct handling of whitespace. In the
    |  #[+src(gh("spaCy", "spacy/tests/conftest.py")) conftest.py], add the new
    |  language ID to the list of #[code _languages]:

+code.
    _languages = ['bn', 'da', 'de', 'en', 'es', 'fi', 'fr', 'he', 'hu', 'it', 'nb',
                  'nl', 'pl', 'pt', 'sv', 'xx'] # new language here

+aside-code("Global tokenizer test example").
    # use fixture by adding it as an argument
    def test_with_all_languages(tokenizer):
        # will be performed on ALL language tokenizers
        tokens = tokenizer(u'Some text here.')

p
    |  The language will now be included in the #[code tokenizer] test fixture,
    |  which is used by the basic tokenizer tests. If you want to add your own
    |  tests that should be run over all languages, you can use this fixture as
    |  an argument of your test function.

+h(3, "testing-custom") Writing language-specific tests

p
    |  It's recommended to always add at least some tests with examples specific
    |  to the language. Language tests should be located in
    |  #[+src(gh("spaCy", "spacy/tests/lang")) tests/lang] in a directory named
    |  after the language ID. You'll also need to create a fixture for your
    |  tokenizer in the #[+src(gh("spaCy", "spacy/tests/conftest.py")) conftest.py].
    |  Always use the #[code get_lang_class()] helper function within the fixture,
    |  instead of importing the class at the top of the file. This will load the
    |  language data only when it's needed. (Otherwise, #[em all data] would be
    |  loaded every time you run a test.)

+code.
    @pytest.fixture
    def en_tokenizer():
        return util.get_lang_class('en').Defaults.create_tokenizer()

p
    |  When adding test cases, always
    |  #[+a(gh("spaCy", "spacy/tests#parameters")) #[code parametrize]] them –
    |  this will make it easier for others to add more test cases without having
    |  to modify the test itself. You can also add parameter tuples, for example,
    |  a test sentence and its expected length, or a list of expected tokens.
    |  Here's an example of an English tokenizer test for combinations of
    |  punctuation and abbreviations:

+code("Example test").
    @pytest.mark.parametrize('text,length', [
        ("The U.S. Army likes Shock and Awe.", 8),
        ("U.N. regulations are not a part of their concern.", 10),
        ("“Isn't it?”", 6)])
    def test_en_tokenizer_handles_punct_abbrev(en_tokenizer, text, length):
        tokens = en_tokenizer(text)
        assert len(tokens) == length

+h(2, "vocabulary") Building the vocabulary

p
    |  spaCy expects that common words will be cached in a
    |  #[+api("vocab") #[code Vocab]] instance. The vocabulary caches lexical
    |  features, and makes it easy to use information from unlabelled text
    |  samples in your models. Specifically, you'll usually want to collect
    |  word frequencies, and train two types of distributional similarity model:
    |  Brown clusters, and word vectors. The Brown clusters are used as features
    |  by linear models, while the word vectors are useful for lexical
    |  similarity models and deep learning.

+h(3, "word-frequencies") Word frequencies

p
    |  To generate the word frequencies from a large, raw corpus, you can use the
    |  #[+src(gh("spacy-dev-resources", "training/word_freqs.py")) word_freqs.py]
    |  script from the spaCy developer resources. Note that your corpus should
    |  not be preprocessed (i.e. you need punctuation for example). The
    |  #[+api("cli#model") #[code model]] command expects a tab-separated word
    |  frequencies file with three columns:

+list("numbers")
    +item The number of times the word occurred in your language sample.
    +item The number of distinct documents the word occurred in.
    +item The word itself.

p
    |  An example word frequencies file could look like this:

+code("es_word_freqs.txt", "text").
    6361109	111	Aunque
    23598543	111	aunque
    10097056	111	claro
    193454	111	aro
    7711123	111	viene
    12812323	111	mal
    23414636	111	momento
    2014580	111	felicidad
    233865	111	repleto
    15527	111	eto
    235565	111	deliciosos
    17259079	111	buena
    71155	111	Anímate
    37705	111	anímate
    33155	111	cuéntanos
    2389171	111	cuál
    961576	111	típico

p
    |  You should make sure you use the spaCy tokenizer for your
    |  language to segment the text for your word frequencies. This will ensure
    |  that the frequencies refer to the same segmentation standards you'll be
    |  using at run-time. For instance, spaCy's English tokenizer segments
    |  "can't" into two tokens. If we segmented the text by whitespace to
    |  produce the frequency counts, we'll have incorrect frequency counts for
    |  the tokens "ca" and "n't".

+h(3, "brown-clusters") Training the Brown clusters

p
    |  spaCy's tagger, parser and entity recognizer are designed to use
    |  distributional similarity features provided by the
    |  #[+a("https://github.com/percyliang/brown-cluster") Brown clustering algorithm].
    |  You should train a model with between 500 and 1000 clusters. A minimum
    |  frequency threshold of 10 usually works well.

p
    |  An example clusters file could look like this:

+code("es_clusters.data", "text").
    0000	Vestigial	1
    0000	Vesturland	1
    0000	Veyreau	1
    0000	Veynes	1
    0000	Vexilografía	1
    0000	Vetrigne	1
    0000	Vetónica	1
    0000	Asunden	1
    0000	Villalambrús	1
    0000	Vichuquén	1
    0000	Vichtis	1
    0000	Vichigasta	1
    0000	VAAH	1
    0000	Viciebsk	1
    0000	Vicovaro	1
    0000	Villardeveyo	1
    0000	Vidala	1
    0000	Videoguard	1
    0000	Vedás	1
    0000	Videocomunicado	1
    0000	VideoCrypt	1

+h(3, "word-vectors") Training the word vectors

p
    |  #[+a("https://en.wikipedia.org/wiki/Word2vec") Word2vec] and related
    |  algorithms let you train useful word similarity models from unlabelled
    |  text. This is a key part of using
    |  #[+a("/docs/usage/deep-learning") deep learning] for NLP with limited
    |  labelled data. The vectors are also useful by themselves – they power
    |  the #[code .similarity()] methods in spaCy. For best results, you should
    |  pre-process the text with spaCy before training the Word2vec model. This
    |  ensures your tokenization will match.

p
    | You can use our
    |  #[+src(gh("spacy-dev-resources", "training/word_vectors.py")) word vectors training script],
    |  which pre-processes the text with your language-specific tokenizer and
    |  trains the model using #[+a("https://radimrehurek.com/gensim/") Gensim].
    |  The #[code vectors.bin] file should consist of one word and vector per line.

//-+aside-code("your_data_directory", "yaml").
    ├── vocab/
    |   ├── lexemes.bin
    |   ├── strings.json
    |   └── oov_prob
    ├── pos/
    |   ├── model
    |   └── config.json
    ├── deps/
    |   ├── model
    |   └── config.json
    └── ner/
        ├── model
        └── config.json

+h(2, "train-tagger-parser") Training the tagger and parser

p
    |  You can now train the model using a corpus for your language annotated
    |  with #[+a("http://universaldependencies.org/") Universal Dependencies].
    |  If your corpus uses the
    |  #[+a("http://universaldependencies.org/docs/format.html") CoNLL-U] format,
    |  i.e. files with the extension #[code .conllu], you can use the
    |  #[+api("cli#convert") #[code convert]] command to convert it to spaCy's
    |  #[+a("/docs/api/annotation#json-input") JSON format] for training.

p
    |  Once you have your UD corpus transformed into JSON, you can train your
    |  model use the using spaCy's #[+api("cli#train") #[code train]] command:

+code(false, "bash").
    python -m spacy train [lang] [output_dir] [train_data] [dev_data] [--n-iter] [--n-sents] [--use-gpu] [--no-tagger] [--no-parser] [--no-entities]