MetaMapLite: A lighter named-entity recognizer

The primary goal of MetaMapLite to provide a near real-time named-entity recognizer which is not a rigorous as MetaMap but much faster while allowing users to customize and augment its behavior for specific purposes.

It uses some of the tables used by MetaMap but all lexical variants used in the table are pre-processed. Named Entities are found using longest match. Restriction by UMLS source and Semantic type is optional. Part-of-speech tagging which improves precision by a small amount (at the cost of speed) is also optional. Negation detection is available using either Wendy Chapman’s context or a native negation detection algorithm based on Wendy Chapman’s NegEx which is somewhat less effective, but faster.

It has:

What is missing:


For running

For Development

Command Line Usage

Example of invocation using script:

./ [options] [<input file>|--]

Example of invocation using Java VM directly:

$ java -cp public_mm_lite/target/metamaplite-3.6-standalone.jar \
      gov.nih.nlm.nls.ner.MetaMapLite \
      --indexdir=public_mm_lite/data/ivf/strict \
      --modelsdir=public_mm_lite/data/models \
      --specialtermsfile=public_mm_lite/data/specialterms.txt  [options] [<input file>|--]

Current options are:

input options:

--                             Read from standard input
--pipe                         Read from standard input

Configuration Options:

--configfile=<filename>        Use configuration file
--set_property=name=value      set property "name" to value

--filelistfn=<filename>        file containing a list of files to processed, one line per file
--filelist=<file0,file1,...>   list of files to processed separated by commas

Options that can be used to override configuration file or when configuration file is not present:

--indexdir=<directory>         location of program's index directory
--modelsdir=<directory>        location of models for sentence breaker and part-of-speech tagger
--specialtermsfile=<filename>  location of file of terms to be excluded

document processing options:

--freetext      Text with no markup.
--ncbicorpus    NCBI Disease Corpus: tab separated fields: id \t title \t abstract
--chemdner      CHEMDNER document: tab separated fields: id \t title \t abstract
--chemdnersldi  CHEMDNER document: id with pipe followed by tab separated fields: id |t title \t abstract
                Use input format specified by loader name.

output options:

--bioc|cdi|bc|bc-evaluate   output compatible with evaluation program bc-evaluate
--mmilike|mmi               similar to MetaMap Fielded MMI output
--brat                      BRAT annotation format

processing options:

                       Set method for text segmentation
--segment_sentences    Segment text by sentence
--segment_blanklines   Segment text by blankline
--segment_lines        Segment text by line
--usecontext           Use ConText Negation Detector instead of NLM's implementation of NegEx
                       Use a user-defined class for negation detector, class must implement to
                       gov.nih.nlm.nls.metamap.lite.NegationDetector interface.

alternate output options:

--list_sentences          list sentences in input
--list_acronyms           list acronyms in input if present.
--list_sentences_postags  list sentences in input with part-of-speech tags


Command line and System properties for metamaplite

These properties can be set using a System property (-D{propertyname}={value}).

|             | load configuration from file (default: ./config/

These properties can be set using a System property (-D{propertyname}={value}) or in configuration file.

| metamaplite.document.inputtype        | document input type (default: freetext)
| metamaplite.outputextension           | result output file extension (default: .mmi)
| metamaplite.outputformat              | result output format (default: mmi)

Processing properties

| metamaplite.segmentation.method       | Set method for text segmentation (values: SENTENCES, BLANKLINES, LINES; default: SENTENCES)
| metamaplite.sourceset                 | use only concepts from listed sources (default: all)
| metamaplite.semanticgroup             | use only concepts belonging to listed semantic types (default: all)
| metamaplite.negation.detector         | negation detector class: default: gov.nih.nlm.nls.metamap.lite.NegEx
| metamaplite.normalized.string.cache.size | set maximum size of string -> normalized string cache
| metamaplite.normalized.string.cache.enable | if true enable string -> normalized string cache
| metamaplite.entitylookup4.term.concept.cache.enable | if true enable term -> concept info cache
| metamaplite.entitylookup4.term.concept.cache.size | set maximum size of term -> concept info cache
| metamaplite.entitylookup4.cui.preferredname.cache.enable |  if true enable cui -> preferred name cache
| metamaplite.entitylookup4.cui.preferredname.cache.size | set maximum size cui -> preferred name cache

Configuration properties

| metamaplite.excluded.termsfile        | cui/terms pairs that are exclude from results (default: data/specialterms.txt)
|           | the directory the indexes resides (sets the following properties)
| metamaplite.ivf.cuiconceptindex       | cui/concept/preferredname index
| metamaplite.ivf.cuisourceinfoindex    | cui/sourceinfo index 
| metamaplite.ivf.cuisemantictypeindex  | cui/semantictype index

|              | the directory the models resides (sets the following properties. default: data/models)
| opennlp.en-pos.bin.path               | (default: data/models/en-pos-maxent.bin)
| opennlp.en-token.bin.path             | (default: data/models/en-token.bin)
| opennlp.en-sent.bin.path              | (default: data/models/en-sent.bin)
| metamaplite.enable.postagging         | Enable part of speech tagging (default: "true" [on])

Using MetaMapLite from Java

Creating properties for configuring MetaMapLite Instance:

Properties myProperties = new Properties();

Loading properties file in “config”:

myProperties.load(new FileReader("config/"));

Creating a metamap lite instance:

MetaMapLite metaMapLiteInst = new MetaMapLite(myProperties);

Creating a document list with one or more documents:

BioCDocument document = FreeText.instantiateBioCDocument("diabetes");
List<BioCDocument> documentList = new ArrayList<BioCDocument>();

Getting a list of entities for the document list:

List<Entity> entityList = metaMapLiteInst.processDocumentList(documentList);

Traversing the entity list displaying cui and matching text:

List<Entity> entityList = metaMapLiteInst.processDocumentList(documentList);
for (Entity entity: entityList) {
  for (Ev ev: entity.getEvSet()) {
    System.out.print(ev.getConceptInfo().getCUI() + "|" + entity.getMatchedText());

Processing Single Terms (without periods)

Disable the Part of Speech Tagger using the following property: “metamaplite.enable.postagging=false”. Add the following line right before instantiating the MetaMapLite instance.

myProperties.setProperty("metamaplite.enable.postagging", "false");
MetaMapLite metaMapLiteInst = new MetaMapLite(myProperties);

Add each term as a single document:

BioCDocument document = FreeText.instantiateBioCDocument(term);

Adding MetaMapLite to a webapp (servlet).

WebApp Local Configuration

A extensive example of providing a servlet complete with data and configuration files in the war (web archive) file is available on the MetaMap website on the MetaMapLite web page (

Alternate Configuration

Below is an alternate configuration for users who don’t want to place the configuration and data in webapp deployment archive file (war).

Place the “” file in the tomcat “conf/” directory and specify that in servlet:

public class SampleWebApp extends HttpServlet {
  /** location of configuration file */
  static String configPropertyFilename =
    System.getProperty("", "conf/");
  Properties properties;
  MetaMapLite metaMapLiteInst;

  public SampleWebApp() {
    try { = new Properties();
      // default properties that can be overriden"metamaplite.ivf.cuiconceptindex","data/ivf/strict/indices/cuiconcept");
      // load user properties FileReader(configPropertyFilename));
      this.metaMapLiteInst = new MetaMapLite(;
    } catch (Exception e) {
      throw new RuntimeException(e);

The absolute locations of indexes and model files can be specified in “”.

Using Maven

Installing metamaplite and dependencies into local Maven repository

From public_mm_lite directory install Context, BioC, and NLS NLP libraries

$ mvn install:install-file \
     -Dfile=lib/context-2012.jar \
     -DgroupId=context \
     -DartifactId=context \
     -Dversion=2012 \

$ mvn install:install-file \
     -Dfile=lib/bioc-1.0.1.jar \
     -DgroupId=bioc \
     -DartifactId=bioc \
     -Dversion=1.0.1 \

$ mvn install:install-file \
     -Dfile=lib/nlp-2.4.C.jar \
     -DgroupId=gov.nih.nlm.nls \
     -DartifactId=nlp \
     -Dversion=2.4.C \

Then install metamaplite into your local Maven repository:

$ mvn install

Add metamaplite dependency to POM file

Add the following dependency to your webapps pom.xml:


irutils indexes

Tables and Indexes

Currently, three tables are used:

New indexes used for MetaMap-like scoring and MMI ranked output

Two new indexes have been introduced to support scoring similar to the original MetaMap and MMI ranking of which MetaMap scoring is a component.

NOTE: Currently, the only mechanism for generating the treecodes and vars (variants) tables from a UMLS subset (generated by Metamorphosys) is by installing the original MetaMap and the Data File Builder using the Data Builder to generate the necessary Treecodes and Vars table files. See the next section for information on adding these indices to a custom dataset.

Generating indexes from UMLS tables

The CreateIndexes class generates tables cuiconcept, cuisourceinfo, and cuist from MRCONSO.RRF and MRSTY.RRF and then produces corresponding indexes for tables. If the variants file, vars.txt, and the treecodes file, mesh_tc_relaxed.txt, are present in the directory {ivfdir}/tables then those files will be indexed as well.


 java -Xmx5g -cp target/metamaplite-<version>-standalone.jar \
  gov.nih.nlm.nls.metamap.dfbuilder.CreateIndexes <mrconsofile> <mrstyfile> <ivfdir>

The resulting indices are in /indices. The tables the indexes are generated from are in /tables.

Checking newly generated indexes

You can use the class irutils.MappedMultiKeyIndexLookup to check the new indexes:

 java -Xmx20g -cp target/metamaplite-<version>-standalone.jar \
  irutils.MappedMultiKeyIndexLookup lookup workingdir indexname column query

For example:

 java -Xmx20g -cp target/metamaplite-<version>-standalone.jar \
  irutils.MappedMultiKeyIndexLookup lookup data/ivf/2016AB/USAbase/strict cuisourceinfo 3 heart

Using newly generated indexes with MetaMapLite

To use the new indexes do one of the following:

Use the –indexdir= option:

java -cp target/metamaplite-<version>-standalone.jar \
 gov.nih.nlm.nls.ner.MetaMapLite --indexdir=<ivfdir> <other-options> <other-args>

Or modify the configuration file config/

metamaplite.ivf.cuiconceptindex: <ivfdir>/indices/cuiconcept
mmetamaplite.ivf.cuisourceinfoindex: <ivfdir>/indices/cuisourceinfo
metamaplite.ivf.cuisemantictypeindex: <ivfdir>/indices/cuist

Adding custom input document formats

New document loader class must conform to BioCDocumentLoader interface. One can add a document loader class in MetaMapLite’s classpath to MetaLite’s list of document loaders by adding it to the properties using System properties or modifying MetaMapLite’s configuration file:

Set as system property:

-Dbioc.document.loader.<name>=<fully-specified class name>

For example creating a loader with the name “qadocument”:


Or add it to config/

bioc.document.loader.qadocument: gov.nih.nlm.nls.metamap.document.QAKeyValueDocument

Adding custom result output formats

New result formatter class must conform to ResultFormatter interface. One can add the result formatter to MetaMapLite by adding its class file to MetaMapLite’s classpath and then adding a reference to it as a property:

Set as system property:

-Dmetamaplite.result.formatter.<name>=<fully-specified class name>

For example creating a formatter with the name “bratsemtype”:


Or add it to config/

 metamaplite.result.formatter.brat: examples.BratSemType

Source code for the BratSemType result formatter is provided in the directory public_mm_lite/src/main/java/examples/

A BioC XML to A BioC XML implementation of MetaMapLite

The class gov.nih.nlm.nls.metamap.lite.BioCProcess allows MetaMapLite to process BioC XML input and write the results in BioC XML.

Using from Java

Below is an example using BioC Processing with MetaMapLite

// Initialize MetaMapLite
Properties defaultConfiguration = getDefaultConfiguration();
String configPropertiesFilename =
Properties configProperties = new Properties();
// set any in-line properties here.
configProperties.load(new FileReader(configPropertiesFilename));
Properties properties =
BioCProcess process = new BioCProcess(properties);

// read BioC XML collection
Reader inputReader = new FileReader(inputFile);
BioCFactory bioCFactory = BioCFactory.newFactory("STANDARD");
BioCCollectionReader collectionReader =
BioCCollection collection = collectionReader.readCollection();

// Run named entity recognition on collection
BioCCollection newCollection = process.processCollection(collection);

// write out the annotated collection
File outputFile = new File(outputFilename);
Writer outputWriter = new PrintWriter(outputFile, "UTF-8");
BioCCollectionWriter collectionWriter = bioCFactory.createBioCCollectionWriter(outputWriter);

This process attempts to preserve any annotation present in the original BioC XML input. Some annotations applied to BioC structures before entity lookup, including `tokenization and part-of-speech-tagging, are discarded before the writing out the final annotated collection. This occurs in the entity lookup class BioCEntityLookup (java package: gov.nih.nlm.nls.metamap.lite).

Command Line Usage:

The file is a wrapper for the BioC to BioC pipeline.

./ <bioc-xml-input-file> <bioc-xml-output-file>

Note that the class is missing the command line options handler that is present in the gov.nih.nlm.nls.ner.MetaMapLite class, use the config/ file to set any custom properties or place the properties in another custom file and use the system property “metamaplite.propertyfile” to refer to the custom file.

Omissions in BioC version

The current version of the BioCProcess class does not call the abbreviation detector or the negation detector. This should be relatively simple to add (particularly the abbreviation detector) and will probably be added in the next release.