nametagger_options {nametagger}

Define text transformations serving as predictive elements in the nametagger model

Description

Define text transformations which are relevant in predicting your entity. Typical text transformations are the token itself, the lemma, the parts of speech tag of the token or the token/lemma's and parts of speech tags in the neighbourhood of the word.

Each argument should be a list with elements use and window.

• use is a logical indicating if the transformation should be used in the model.

• window specifies how many adjacent words can observe the feature template value of a given word. The default value of 0 denotes only the word in question, no surrounding words.

If you specifiy the argument without specifying use, it will by default use it. For arguments brown, gazetteers and gazetteers_enhanced, see the examples and the documentation at https://ufal.mff.cuni.cz/nametag/1.

Usage

nametagger_options(
  file = "nametagger.ner",
  type = c("generic", "english", "czech"),
  tagger = c("trivial", "external"),
  token = list(use = FALSE, window = 1),
  token_capitalised = list(use = FALSE, window = 0),
  token_normalised = list(use = FALSE, window = 0),
  token_normalisedsuffix = list(use = FALSE, window = 0, from = 1, to = 4),
  lemma = list(use = FALSE, window = 0),
  lemma_capitalised = list(use = FALSE, window = 0),
  lemma_normalised = list(use = FALSE, window = 0),
  lemma_normalisedsuffix = list(use = FALSE, window = 0, from = 1, to = 4),
  pos = list(use = tagger == "external", window = 0),
  time = list(use = FALSE, window = 0),
  url_email = list(use = FALSE, url = "URL", email = "EMAIL"),
  ner_previous = list(use = FALSE, window = 0),
  brown = list(use = FALSE, window = 0),
  gazetteers = list(use = FALSE, window = 0),
  gazetteers_enhanced = list(use = FALSE)
)

Arguments

file

path to the filename where the model will be saved

type

either one of 'generic', 'english' or 'czech'. See the documentation at the documentation at https://ufal.mff.cuni.cz/nametag/1.

tagger

either one of 'trivial' (no lemma used in the training data), 'external' (you provided your own lemma in the training data)

token

use forms as features

token_capitalised

use capitalization of form as features

token_normalised

use case normalized (first character as-is, others lowercased) forms as features

token_normalisedsuffix

shortest longest – use suffixes of case normalized (first character as-is, others lowercased) forms of lengths between shortest and longest

lemma

use raw lemmas as features

lemma_capitalised

use capitalization of raw lemma as features

lemma_normalised

use case normalized (first character as-is, others lowercased) raw lemmas as features

lemma_normalisedsuffix

shortest longest – use suffixes of case normalized (first character as-is, others lowercased) raw lemmas of lengths between shortest and longest

pos

use parts-of-speech tags as features

time

recognize numbers which could represent hours, minutes, hour:minute time, days, months or years

url_email

If an URL or an email is detected, it is immediately marked with specified named entity type and not used in further processing. The specified entity label to use can be specified with url and email (in that sequence)

ner_previous

use named entities predicted by previous stage as features

brown

file [prefix_lengths] – use Brown clusters found in the specified file. An optional list of lengths of cluster prefixes to be used in addition to the full Brown cluster can be specified.

gazetteers

[files] – use given files as gazetteers. Each file is one gazetteers list independent of the others and must contain a set of lemma sequences, each on a line, represented as raw lemmas separated by spaces.

gazetteers_enhanced

(form|rawlemma|rawlemmas) (embed_in_model|out_of_model) file_base entity [file_base entity ...] – use gazetteers from given files. Each gazetteer contains (possibly multiword) named entities per line. Matching of the named entities can be performed either using form, disambiguated rawlemma of any of rawlemmas proposed by the morphological analyzer. The gazetteers might be embedded in the model file or not; in either case, additional gazetteers are loaded during each startup. For each file_base specified in GazetteersEnhanced templates, three files are tried:

• file_base.txt: gazetteers used as features, representing each file_base with a unique feature

• file_base.hard_pre.txt: matched named entities (finding non-overlapping entities, preferring the ones starting earlier and longer ones in case of ties) are forced to the specified entity type even before the NER model is executed

• file_base.hard_post.txt: after running the NER model, tokens not recognized as entities are matched against the gazetteers (again finding non-overlapping entities, preferring the ones starting earlier and longer ones in case of ties) and marked as entity type if found

Value

an object of class nametagger_options with transformation information to be used by nametagger

Examples

opts <- nametagger_options(token = list(window = 2))
opts
opts <- nametagger_options(time = list(use = TRUE, window = 3),
                           token_capitalised = list(use = TRUE, window = 1),
                           ner_previous = list(use = TRUE, window = 5))
opts                            
opts <- nametagger_options(
  lemma_capitalised = list(window = 3),
  brown = list(window = 1, file = "path/to/brown/clusters/file.txt"),
  gazetteers = list(window = 1, 
                    file_loc = "path/to/txt/file1.txt", 
                    file_time = "path/to/txt/file2.txt"))
opts
opts <- nametagger_options(
  lemma_capitalised = list(window = 3),
  brown = list(window = 2, 
               file = "path/to/brown/clusters/file.txt"),
  gazetteers_enhanced = list(
    loc  = "LOC",  type_loc  = "form", save_loc  = "embed_in_model", file_loc  = "pathto/loc.txt",  
    time = "TIME", type_time = "form", save_time = "embed_in_model", file_time = "pathto/time.txt")
    )
opts