Demonstrations
From LEARN
Contents |
F-EXT-WS: Natural Language Processor
F-EXT-WS is a service that freely provides natural language processors in the Web.
F-EXT-WS input is raw text.
F-EXT-WS output is the given input text along with syntactic and shallow semantic annotation.
Currently, the following multilingual annotation tasks are provided by the F-EXT-WS.
- Portuguese Language
- Part-of-speech (POS) tagging
- Base noun phrase chunking
- Phrase chunking: noun, verbal and prepositional
- Named entity recognition
- Clause identification
- English Language
- Part-of-speech (POS) tagging
- Phrase chunking
- Clause identification
- Semantic role labeling
F-EXT-WS-2.0: A Web Service for Natural Language Processing
F-EXT-WS-2.0 exposes the natural language processors through a web service.
The service WSDL is available at http://baixo.learn.fplf.org.br/axis/services/FEXT20?wsdl
In order to use the web service interface, one must register at the F-EXT-WS web site.
The web service is composed by three functions:
- xsd:int runTask (username, password, language, linguisticinfo, inText)
Input parameters
A valid username and corresponding password must be provided to run a task.
language must be either 'por' for Portuguese processing or 'eng' for English.
linguisticinfo specifies the output required and must be provided the following way:
For Portuguese, available values are:
'pos' - Part-of-speech
'np' - Noun phrase chunking
'ner' - Named entity recognition
'ck' - Text chunking
'clause' - Clause identification
For English, available values are:
'pos' - Part-of-speech
'np' - Noun phrase chunking
'ck' - Text chunking
'cl' - Clause identification
'srl'- Semantic role labeling
It is possible to combine multiple outputs using # as separator, e.g., linguisticinfo='pos#np' will provide both part-of-speech and noun phrase tags as output.
Output
runTask returns either a positive integer (taskID) that identifies the task or a negative value (errorCode) that indicates an error (see getErrorMessage function below).
- xsd:string getResult (username, password, taskID)
After calling runTask, getResult is called to retrieve the processing results.
Results are provided as a tabular output, like in the example below:
[features = word, pos, np] O ART I Brasil NPROP I leva V O susto N I de PREP I os ART I EUA NPROP I
- xsd:string getErrorMessage (errorCode)
In case runTask returns a negative value (error code), getErrorMessage may be called to retrieve the error message.
Source Code
We have made available the source code of some computational linguistic tools.
Large Margin Structured Perceptron
Structured learning consists in learning a mapping from inputs to structured outputs by means of a sample of correct input-output pairs. Many important problems fit into this setting. For instance, part-of-speech tagging consists of finding a sequence of tags for a given sentence, dependency parsing involves the recognition of a tree structure underlying a sentence, and coreference resolution is to cluster a given set of references within a document.
Structured perceptron is an online algorithm to train a linear predictor for structured problems. This project consists of a structured learning framework based on structured perceptrons and also includes some instantiations for computational linguistic problems.
