Learn R Programming

NLP (version 0.3-0)

annotators: Simple annotator generators

Description

Create annotator objects for composite basic NLP tasks based on functions performing simple basic tasks.

Usage

Simple_Para_Token_Annotator(f, meta = list(), classes = NULL)
Simple_Sent_Token_Annotator(f, meta = list(), classes = NULL)
Simple_Word_Token_Annotator(f, meta = list(), classes = NULL)
Simple_POS_Tag_Annotator(f, meta = list(), classes = NULL)
Simple_Entity_Annotator(f, meta = list(), classes = NULL)
Simple_Chunk_Annotator(f, meta = list(), classes = NULL)
Simple_Stem_Annotator(f, meta = list(), classes = NULL)

Value

An annotator object inheriting from the given classes and the default ones.

Arguments

f

a function performing a “simple” basic NLP task (see Details).

meta

an empty or named list of annotator (pipeline) metadata tag-value pairs.

classes

a character vector or NULL (default) giving classes to be used for the created annotator object in addition to the default ones (see Details).

Details

The purpose of these functions is to facilitate the creation of annotators for basic NLP tasks as described below.

Simple_Para_Token_Annotator() creates “simple” paragraph token annotators. Argument f should be a paragraph tokenizer, which takes a string s with the whole text to be processed, and returns the spans of the paragraphs in s, or an annotation object with these spans and (possibly) additional features. The generated annotator inherits from the default classes "Simple_Para_Token_Annotator" and "Annotator". It uses the results of the simple paragraph tokenizer to create and return annotations with unique ids and type ‘paragraph’.

Simple_Sent_Token_Annotator() creates “simple” sentence token annotators. Argument f should be a sentence tokenizer, which takes a string s with the whole text to be processed, and returns the spans of the sentences in s, or an annotation object with these spans and (possibly) additional features. The generated annotator inherits from the default classes "Simple_Sent_Token_Annotator" and "Annotator". It uses the results of the simple sentence tokenizer to create and return annotations with unique ids and type ‘sentence’, possibly combined with sentence constituent features for already available paragraph annotations.

Simple_Word_Token_Annotator() creates “simple” word token annotators. Argument f should be a simple word tokenizer, which takes a string s giving a sentence to be processed, and returns the spans of the word tokens in s, or an annotation object with these spans and (possibly) additional features. The generated annotator inherits from the default classes "Simple_Word_Token_Annotator" and "Annotator". It uses already available sentence token annotations to extract the sentences and obtains the results of the word tokenizer for these. It then adds the sentence character offsets and unique word token ids, and word token constituents features for the sentences, and returns the word token annotations combined with the augmented sentence token annotations.

Simple_POS_Tag_Annotator() creates “simple” POS tag annotators. Argument f should be a simple POS tagger, which takes a character vector giving the word tokens in a sentence, and returns either a character vector with the tags, or a list of feature maps with the tags as ‘POS’ feature and possibly other features. The generated annotator inherits from the default classes "Simple_POS_Tag_Annotator" and "Annotator". It uses already available sentence and word token annotations to extract the word tokens for each sentence and obtains the results of the simple POS tagger for these, and returns annotations for the word tokens with the features obtained from the POS tagger.

Simple_Entity_Annotator() creates “simple” entity annotators. Argument f should be a simple entity detector (“named entity recognizer”) which takes a character vector giving the word tokens in a sentence, and return an annotation object with the word token spans, a ‘kind’ feature giving the kind of the entity detected, and possibly other features. The generated annotator inherits from the default classes "Simple_Entity_Annotator" and "Annotator". It uses already available sentence and word token annotations to extract the word tokens for each sentence and obtains the results of the simple entity detector for these, transforms word token spans to character spans and adds unique ids, and returns the combined entity annotations.

Simple_Chunk_Annotator() creates “simple” chunk annotators. Argument f should be a simple chunker, which takes as arguments character vectors giving the word tokens and the corresponding POS tags, and returns either a character vector with the chunk tags, or a list of feature lists with the tags as ‘chunk_tag’ feature and possibly other features. The generated annotator inherits from the default classes "Simple_Chunk_Annotator" and "Annotator". It uses already available annotations to extract the word tokens and POS tags for each sentence and obtains the results of the simple chunker for these, and returns word token annotations with the chunk features (only).

Simple_Stem_Annotator() creates “simple” stem annotators. Argument f should be a simple stemmer, which takes as arguments a character vector giving the word tokens, and returns a character vector with the corresponding word stems. The generated annotator inherits from the default classes "Simple_Stem_Annotator" and "Annotator". It uses already available annotations to extract the word tokens, and returns word token annotations with the corresponding stem features (only).

In all cases, if the underlying simple processing function returns annotation objects these should not provide their own ids (or use such in the features), as the generated annotators will necessarily provide these (the already available annotations are only available at the annotator level, but not at the simple processing level).

See Also

Package openNLP which provides annotator generators for sentence and word tokens, POS tags, entities and chunks, using processing functions based on the respective Apache OpenNLP MaxEnt processing resources.

Examples

Run this code
## A simple text.
s <- String("  First sentence.  Second sentence.  ")
##           ****5****0****5****0****5****0****5**

## A very trivial sentence tokenizer.
sent_tokenizer <-
function(s) {
    s <- as.String(s)
    m <- gregexpr("[^[:space:]][^.]*\\.", s)[[1L]]
    Span(m, m + attr(m, "match.length") - 1L)
}
## (Could also use Regexp_Tokenizer() with the above regexp pattern.)
sent_tokenizer(s)
## A simple sentence token annotator based on the sentence tokenizer.
sent_token_annotator <- Simple_Sent_Token_Annotator(sent_tokenizer)
sent_token_annotator
a1 <- annotate(s, sent_token_annotator)
a1
## Extract the sentence tokens.
s[a1]

## A very trivial word tokenizer.
word_tokenizer <-
function(s) {
    s <- as.String(s)
    ## Remove the last character (should be a period when using
    ## sentences determined with the trivial sentence tokenizer).
    s <- substring(s, 1L, nchar(s) - 1L)
    ## Split on whitespace separators.
    m <- gregexpr("[^[:space:]]+", s)[[1L]]
    Span(m, m + attr(m, "match.length") - 1L)
}
lapply(s[a1], word_tokenizer)
## A simple word token annotator based on the word tokenizer.
word_token_annotator <- Simple_Word_Token_Annotator(word_tokenizer)
word_token_annotator
a2 <- annotate(s, word_token_annotator, a1)
a2
## Extract the word tokens.
s[subset(a2, type == "word")]

## A simple word token annotator based on wordpunct_tokenizer():
word_token_annotator <-
    Simple_Word_Token_Annotator(wordpunct_tokenizer,
                                list(description =
                                     "Based on wordpunct_tokenizer()."))
word_token_annotator
a2 <- annotate(s, word_token_annotator, a1)
a2
## Extract the word tokens.
s[subset(a2, type == "word")]

Run the code above in your browser using DataLab