Skip to content

plandes/clj-nlp-parse

Repository files navigation

Natural Language Parse and Feature Generation

Travis CI Build Status

This repository provides generalized library to deal with natural language. Specifically it:

  • Wraps several Java natural language parsing libraries.
  • Gives access the data structures rendered by the parsers.
  • Provides utility functions to create features.

This framework combines the results of the following frameworks:

Table of Contents

Features

Obtaining

In your project.clj file, add:

Clojars Project

Documentation

API Documentation

Annotation Definitions

The utterance parse annotation tree definitions is given here.

Example Parse

An example of a full annotation parse is given here.

Setup

The NER model is included in the Stanford CoreNLP dependencies, but you still have to download the POS model. To download (or create a symbolic link if you've set the ZMODEL environment variable):

$ make model

If this doesn't work, follow the manual steps. Otherwise you can optionally move the model to a shared location on the file system and skip to configuring the REPL.

Download and Install POS Tagger Model Manually

If the normal setup failed, you'll have to manually download the POS tagger model.

The library can be configured to use any POS model (or NER for that matter), but by default it expects the english-left3words-distsim.tagger model.

  1. Create a directory where to put the model

    $ mkdir -p path-to-model/stanford/pos
  2. Download the english-left3words-distsim.tagger model the or similar model.

  3. Install the model file:

    $ unzip stanford-postagger-2015-12-09.zip
    $ mv stanford-postagger-2015-12-09/models/english-left3words-distsim.tagger path-to-model/stanford/pos

REPL

If you download the model in to any other location other that the current start directory (see setup) you will have to tell the REPL where the model is kept on the file system.

Start the REPL and configure:

user> (System/setProperty "zensols.model" "path-to-model")

Note that system properties can be passed via lein to avoid having to repeat this for each REPL instance.

Usage

This package supports:

Usage Example

See the example repo that illustrates how to use this library and contains the code from where these examples originate. It's highly recommended to clone it and follow along as you peruse this README.

Parsing an Utterance

user> (require '[zensols.nlparse.parse :refer (parse)])
user> (clojure.pprint/pprint (parse "I am Paul Landes."))
=> {:text "I am Paul Landes.",
 :mentions
 ({:entity-type "PERSON",
   :token-range [2 4],
   :ner-tag "PERSON",
   :sent-index 0,
   :char-range [5 16],
   :text "Paul Landes"}),
 :sents
 ({:text "I am Paul Landes.",
   :sent-index 0,
   :parse-tree
   {:label "ROOT",
    :child
    ({:label "S",
      :child
      ({:label "NP",
        :child ({:label "PRP", :child ({:label "I", :token-index 1})})}
...
   :dependency-parse-tree
   ({:token-index 4,
     :text "Landes",
     :child
     ({:dep "nsubj", :token-index 1, :text "I"}
      {:dep "cop", :token-index 2, :text "am"}
      {:dep "compound", :token-index 3, :text "Paul"}
      {:dep "punct", :token-index 5, :text "."})}),
...
   :tokens
   ({:token-range [0 1],
     :ner-tag "O",
     :pos-tag "PRP",
     :lemma "I",
     :token-index 1,
     :sent-index 0,
     :char-range [0 1],
     :text "I",
     :srl
     {:id 1,
      :propbank nil,
      :head-id 2,
      :dependency-label "root",
      :heads ({:function-tag "PPT", :dependency-label "A1"})}}
...

Utility Functions

There utility function to have with getting around the parsed data, as it can be pretty large. For example, to find the head of the dependency head tree:

(def panon (parse "I am Paul Landes."))
=> {:text...
user> (->> panon :sents first p/root-dependency :text)
=> "Landes"

In this case, the last name is the head of tree and happens to be a named entity as detected by the Stanford CoreNLP NER system. Named entities are annotatated at the token level, but also included in the mentions top level with the entire set of concatenated tokens (for cases where an NER contains more than one token like in this case). To get the full mention text:

user> (->> panon :sents first p/root-dependency
                (p/mention-for-token panon)
                first :text))
=> "Paul Landes"

Feature Creation

This library was written to generate features for a machine learning algoritms. There are some utility functions for doing this.

Other feature libraries the integrate with this library:

Below are examples of feature creation with just this library.

Get the first propbank parsed from the SRL:

user> (->> panon f/first-propbank-label)
=> "be.01"

Get stats on features:

user> (->> panon p/tokens (f/token-features panon))
=> {:utterance-length 17,
    :mention-count 1,
	:sent-count 1,
	:token-count 5,
	:token-average-length 14/5,
	:is-question false}

Each function X has an analog function X-feature-keys that describes the features generates and their types, which can be used directly as Weka attributes:

user> (clojure.pprint/pprint (f/token-feature-metas))
=> [[:utterance-length numeric]
    [:mention-count numeric]
	[:sent-count numeric]
	[:token-count numeric]
	[:token-average-length numeric]
	[:is-question boolean]]

Get in/out-of-vocabulary ratio:

user> (->> panon p/tokens f/dictionary-features)
=> {:in-dict-ratio 4/5}

Word count features provide distributions over word counts. See the unit test.

Stopword Filtering

Filter

user> (require '[zensols.nlparse.parse :as p])
user> (require '[zensols.nlparse.stopword :as st])
user> (->> (p/parse "This is a test.  This will filter 5 semantically significant words.")
           p/tokens
           st/go-word-forms)
=> ("test" "filter" "semantically" "significant" "words")

See the unit test.

Dictionary Utility

See the NLP feature library for more information on dictionary specifics.

Pipeline Configuration

You can not only configure the natural language processing pipeline and which specific components to use, but you can also define and add your own plugin library. See the config namespace for more information.

Pipeline Usage

For example, if all you need is tokenization and sentence chunking create a context and parse it using macro with-context and the context you create with specific components:

(require '[zensols.nlparse.config :as conf :refer (with-context)]
         '[zensols.nlparse.parse :refer (parse)])

(let [ctx (->> (conf/create-parse-config
                :pipeline [(conf/tokenize)
                           (conf/sentence)])
               conf/create-context)]
  (with-context ctx
    (parse "I love Clojure.  I enjoy it.")))

You can also specify the configuration in the form of a string:

(let [ctx (conf/create-context "tokenize,sentence,part-of-speech")]
  (with-context ctx
    (parse "I love Clojure.  I enjoy it.")))

The configuration string can also take parameters (ex the en parameter to the tokenizer specifying English as the natural language):

(let [ctx (conf/create-context "tokenize(en),sentence,part-of-speech")]
  (with-context ctx
    (parse "I love Clojure.  I enjoy it.")))

For an example on how to configure the pipeline, see this test case. For more information on the DSL itself see the DSL parser.

Convenience Namespace

If you use a particular configuration that doesn't change often consider your own utility parse namespace:

(ns example.nlp.parse
  (:require [zensols.nlparse.parse :as p]
            [zensols.nlparse.config :as conf :refer (with-context)]))

(defonce ^:private parse-context-inst (atom nil))

(defn- create-context []
  (->> ["tokenize"
        "sentence"
        "part-of-speech"
        "morphology"
        "named-entity-recognizer"
        "parse-tree"]
       (clojure.string/join ",")
       conf/create-context))

(defn- context []
  (swap! parse-context-inst #(or % (create-context))))

(defn parse [utterance]
  (with-context (context)
    (p/parse utterance)))

Now in your application namespace:

(ns example.nlp.core
  (:require [example.nlp.parse :as p]))

(defn somefn []
  (p/parse "an utterance"))

Command Line Usage

The command line usage of this project has moved to the NLP server.

Building

To build from source, do the folling:

  • Install Leiningen (this is just a script)
  • Install GNU make
  • Install Git
  • Download the source: git clone --recurse-submodules https://github.com/plandes/clj-nlp-parse && cd clj-nlp-parse
  • Build the software: make jar
  • Build the distribution binaries: make dist

Note that you can also build a single jar file with all the dependencies with: make uber

Changelog

An extensive changelog is available here.

Citation

If you use this software in your research, please cite with the following BibTeX:

@misc{plandes-clj-nlp-parse,
  author = {Paul Landes},
  title = {Natural Language Parse and Feature Generation},
  year = {2018},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/plandes/clj-nlp-parse}}
}

References

See the General NLP feature creation library for additional references.

@phdthesis{choi2014optimization,
  title = {Optimization of natural language processing components for robustness and scalability},
  author = {Choi, Jinho D},
  year = {2014},
  school = {University of Colorado Boulder}
}

@InProceedings{manning-EtAl:2014:P14-5,
  author = {Manning, Christopher D. and  Surdeanu, Mihai  and  Bauer, John  and  Finkel, Jenny  and  Bethard, Steven J. and  McClosky, David},
  title = {The {Stanford} {CoreNLP} Natural Language Processing Toolkit},
  booktitle = {Association for Computational Linguistics (ACL) System Demonstrations},
  year = {2014},
  pages = {55--60},
  url = {http://www.aclweb.org/anthology/P/P14/P14-5010}
}

License

Copyright (c) 2016, 2017, 2018 Paul Landes

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.