Detecting Languages During Indexing

Solr can identify languages and map text to language-specific fields during indexing using the langid UpdateRequestProcessor.

Solr supports three implementations of this feature:

You can see a comparison between the Tika and LangDetect implementations here: http://blog.mikemccandless.com/2011/10/accuracy-and-performance-of-googles.html. In general, the LangDetect implementation supports more languages with higher performance.

For specific information on each of these language identification implementations, including a list of supported languages for each, see the relevant project websites.

For more information about language analysis in Solr, see Language Analysis.

Configuring Language Detection

You can configure the langid UpdateRequestProcessor in solrconfig.xml. Both implementations take the same parameters, which are described in the following section. At a minimum, you must specify the fields for language identification and a field for the resulting language code.

Configuring Tika Language Detection

Here is an example of a minimal Tika langid configuration in solrconfig.xml:

<processor class="org.apache.solr.update.processor.TikaLanguageIdentifierUpdateProcessorFactory">
  <lst name="defaults">
    <str name="langid.fl">title,subject,text,keywords</str>
    <str name="langid.langField">language_s</str>
  </lst>
</processor>

Configuring LangDetect Language Detection

Here is an example of a minimal LangDetect langid configuration in solrconfig.xml:

<processor class="org.apache.solr.update.processor.LangDetectLanguageIdentifierUpdateProcessorFactory">
  <lst name="defaults">
    <str name="langid.fl">title,subject,text,keywords</str>
    <str name="langid.langField">language_s</str>
  </lst>
</processor>

Configuring OpenNLP Language Detection

Here is an example of a minimal OpenNLP langid configuration in solrconfig.xml:

<processor class="org.apache.solr.update.processor.OpenNLPLangDetectUpdateProcessorFactory">
  <lst name="defaults">
    <str name="langid.fl">title,subject,text,keywords</str>
    <str name="langid.langField">language_s</str>
    <str name="langid.model">langdetect-183.bin</str>
  </lst>
</processor>

OpenNLP-specific Parameters

langid.model

An OpenNLP language detection model. The OpenNLP project provides a pre-trained 103 language model on the OpenNLP site’s model dowload page. Model training instructions are provided on the OpenNLP website. This parameter is required. See Resource and Plugin Loading for information on where to put the model.

OpenNLP Language Codes

OpenNLPLangDetectUpdateProcessor automatically converts the 3-letter ISO 639-3 codes detected by the OpenNLP model into 2-letter ISO 639-1 codes.

langid Parameters

As previously mentioned, both implementations of the langid UpdateRequestProcessor take the same parameters.

langid

When true, the default, enables language detection.

langid.fl

A comma- or space-delimited list of fields to be processed by langid. This parameter is required.

langid.langField

Specifies the field for the returned language code. This parameter is required.

langid.langsField

Specifies the field for a list of returned language codes. If you use langid.map.individual, each detected language will be added to this field.

langid.overwrite

Specifies whether the content of the langField and langsField fields will be overwritten if they already contain values. The default is false.

langid.lcmap

A space-separated list specifying colon delimited language code mappings to apply to the detected languages.

For example, you might use this to map Chinese, Japanese, and Korean to a common cjk code, and map both American and British English to a single en code by using langid.lcmap=ja:cjk zh:cjk ko:cjk en_GB:en en_US:en.

This affects both the values put into the langField and langsField fields, as well as the field suffixes when using langid.map, unless overridden by langid.map.lcmap.

langid.threshold

Specifies a threshold value between 0 and 1 that the language identification score must reach before langid accepts it.

With longer text fields, a high threshold such as 0.8 will give good results. For shorter text fields, you may need to lower the threshold for language identification, though you will be risking somewhat lower quality results. We recommend experimenting with your data to tune your results.

The default is 0.5.

langid.whitelist

Specifies a list of allowed language identification codes. Use this in combination with langid.map to ensure that you only index documents into fields that are in your schema.

langid.map

Enables field name mapping. If true, Solr will map field names for all fields listed in langid.fl. The default is false.

langid.map.fl

A comma-separated list of fields for langid.map that is different than the fields specified in langid.fl.

langid.map.keepOrig

If true, Solr will copy the field during the field name mapping process, leaving the original field in place. The default is false.

langid.map.individual

If true, Solr will detect and map languages for each field individually. The default is false.

langid.map.individual.fl

A comma-separated list of fields for use with langid.map.individual that is different than the fields specified in langid.fl.

langid.fallback

Specifies a language code to use if no language is detected or specified in langid.fallbackFields.

langid.fallbackFields

If no language is detected that meets the langid.threshold score, or if the detected language is not on the langid.whitelist, this field specifies language codes to be used as fallback values.

If no appropriate fallback languages are found, Solr will use the language code specified in langid.fallback.

langid.map.lcmap

A space-separated list specifying colon-delimited language code mappings to use when mapping field names.

For example, you might use this to make Chinese, Japanese, and Korean language fields use a common *_cjk suffix, and map both American and British English fields to a single *_en by using langid.map.lcmap=ja:cjk zh:cjk ko:cjk en_GB:en en_US:en.

A list defined with this parameter will override any configuration set with langid.lcmap.

langid.map.pattern

By default, fields are mapped as <field>_<language>. To change this pattern, you can specify a Java regular expression in this parameter.

langid.map.replace

By default, fields are mapped as <field>_<language>. To change this pattern, you can specify a Java replace in this parameter.

langid.enforceSchema

If false, the langid processor does not validate field names against your schema. This may be useful if you plan to rename or delete fields later in the UpdateChain.

The default is true.