• this library tries to solve language detection of very short words and phrases, even shorter than tweets
• makes use of both statistical and rule-based approaches
• outperforms Apache Tika, Apache OpenNLP and Optimaize Language Detector for more than 70 languages
• works within every Java 6+ application
• no additional training of language models necessary
• api for adding your own language models
• offline usage without having to connect to an external service or API
• can be used in a REPL for a quick try-out

Its task is simple: It tells you which language some provided textual data is written in. This is very useful as a preprocessing step for linguistic data in natural language processing applications such as text classification and spell checking. Other use cases, for instance, might include routing e-mails to the right geographically located customer service department, based on the e-mails' languages.

Language detection is often done as part of large machine learning frameworks or natural language processing applications. In cases where you don't need the full-fledged functionality of those systems or don't want to learn the ropes of those, a small flexible library comes in handy.

So far, three other comprehensive open source libraries working on the JVM for this task are Apache Tika, Apache OpenNLP and Optimaize Language Detector. Unfortunately, especially the latter has three major drawbacks:

1. Detection only works with quite lengthy text fragments. For very short text snippets such as Twitter messages, it doesn't provide adequate results.
2. The more languages take part in the decision process, the less accurate are the detection results.
3. Configuration of the library is quite cumbersome and requires some knowledge about the statistical methods that are used internally.

Lingua aims at eliminating these problems. It nearly doesn't need any configuration and yields pretty accurate results on both long and short text, even on single words and phrases. It draws on both rule-based and statistical methods but does not use any dictionaries of words. It does not need a connection to any external API or service either. Once the library has been downloaded, it can be used completely offline.

Compared to other language detection libraries, Lingua's focus is on quality over quantity, that is, getting detection right for a small set of languages first before adding new ones. Currently, the following 75 languages are supported:

• A
  • Afrikaans
  • Albanian
  • Arabic
  • Armenian
  • Azerbaijani
• B
  • Basque
  • Belarusian
  • Bengali
  • Norwegian Bokmal
  • Bosnian
  • Bulgarian
• C
  • Catalan
  • Chinese
  • Croatian
  • Czech
• D
  • Danish
  • Dutch
• E
  • English
  • Esperanto
  • Estonian
• F
  • Finnish
  • French
• G
  • Ganda
  • Georgian
  • German
  • Greek
  • Gujarati
• H
  • Hebrew
  • Hindi
  • Hungarian
• I
  • Icelandic
  • Indonesian
  • Irish
  • Italian
• J
  • Japanese
• K
  • Kazakh
  • Korean
• L
  • Latin
  • Latvian
  • Lithuanian
• M
  • Macedonian
  • Malay
  • Maori
  • Marathi
  • Mongolian
• N
  • Norwegian Nynorsk
• P
  • Persian
  • Polish
  • Portuguese
  • Punjabi
• R
  • Romanian
  • Russian
• S
  • Serbian
  • Shona
  • Slovak
  • Slovene
  • Somali
  • Sotho
  • Spanish
  • Swahili
  • Swedish
• T
  • Tagalog
  • Tamil
  • Telugu
  • Thai
  • Tsonga
  • Tswana
  • Turkish
• U
  • Ukrainian
  • Urdu
• V
  • Vietnamese
• W
  • Welsh
• X
  • Xhosa
• Y
  • Yoruba
• Z
  • Zulu

Lingua is able to report accuracy statistics for some bundled test data available for each supported language. The test data for each language is split into three parts:

1. a list of single words with a minimum length of 5 characters
2. a list of word pairs with a minimum length of 10 characters
3. a list of complete grammatical sentences of various lengths

Both the language models and the test data have been created from separate documents of the Wortschatz corpora offered by Leipzig University, Germany. Data crawled from various news websites have been used for training, each corpus comprising one million sentences. For testing, corpora made of arbitrarily chosen websites have been used, each comprising ten thousand sentences. From each test corpus, a random unsorted subset of 1000 single words, 1000 word pairs and 1000 sentences has been extracted, respectively.

Given the generated test data, I have compared the detection results of Lingua, Apache Tika, Apache OpenNLP and Optimaize Language Detector using parameterized JUnit tests running over the data of Lingua's supported 75 languages. Languages that are not supported by the other libraries are simply ignored for those during the detection process.

Each of the following sections contains two plots. The bar plot shows the detailed accuracy results for each supported language. The box plot illustrates the distributions of the accuracy values for each classifier. The boxes themselves represent the areas which the middle 50 % of data lie within. Within the colored boxes, the horizontal lines mark the median of the distributions.

Bar plot

Bar plot

Bar plot

Bar plot

The table below shows detailed statistics for each language and classifier including mean, median and standard deviation.

Open table

Language	Average					Single Words					Word Pairs					Sentences
	Lingua (high accuracy mode)	Lingua (low accuracy mode)	Tika	OpenNLP	Optimaize	Lingua (high accuracy mode)	Lingua (low accuracy mode)	Tika	OpenNLP	Optimaize	Lingua (high accuracy mode)	Lingua (low accuracy mode)	Tika	OpenNLP	Optimaize	Lingua (high accuracy mode)	Lingua (low accuracy mode)	Tika	OpenNLP	Optimaize
Afrikaans	79	64	71	72	39	58	38	44	41	3	81	62	70	75	22	97	93	98	99	93
Albanian	88	80	79	71	70	69	54	54	40	38	95	86	84	73	73	100	99	99	100	98
Arabic	98	94	97	84	89	96	88	94	65	72	99	96	99	88	94	100	99	100	99	100
Armenian	100	100	-	100	-	100	100	-	100	-	100	100	-	100	-	100	100	-	100	-
Azerbaijani	90	82	-	82	-	77	71	-	60	-	92	78	-	86	-	99	96	-	99	-
Basque	84	74	83	77	66	71	56	64	56	33	87	76	86	82	70	93	91	98	92	95
Belarusian	97	92	96	91	87	92	80	92	78	69	99	95	98	95	92	100	100	100	100	99
Bengali	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100
Bokmal	58	49	-	66	-	39	27	-	42	-	59	47	-	69	-	75	74	-	87	-
Bosnian	35	29	-	26	-	29	23	-	12	-	35	29	-	22	-	40	36	-	44	-
Bulgarian	87	78	73	83	48	70	56	52	62	18	91	81	69	87	36	99	96	96	100	91
Catalan	70	58	58	42	31	51	33	32	11	2	74	60	57	32	16	86	81	84	81	77
Chinese	100	100	69	78	31	100	100	20	40	0	100	100	86	94	2	100	100	100	100	91
Croatian	72	60	74	50	41	53	36	54	23	8	74	57	72	44	24	90	85	97	81	91
Czech	80	71	72	67	49	66	54	54	42	21	84	72	75	70	46	91	87	88	90	81
Danish	81	70	83	60	55	61	45	63	34	19	84	70	86	52	51	98	95	99	94	96
Dutch	77	64	60	61	39	55	36	31	31	6	81	61	52	57	19	96	94	98	97	91
English	81	62	64	52	41	55	29	30	10	2	89	62	62	46	23	99	96	99	99	97
Esperanto	84	66	-	76	-	67	44	-	50	-	85	61	-	78	-	98	92	-	100	-
Estonian	92	83	84	59	61	80	62	66	29	23	96	88	88	60	63	100	99	100	88	98
Finnish	96	91	94	86	79	90	77	86	68	51	98	95	96	91	86	100	100	100	100	100
French	89	77	78	59	54	74	52	55	25	18	94	83	80	55	48	99	97	99	98	97
Ganda	91	84	-	-	-	79	65	-	-	-	95	87	-	-	-	100	100	-	-	-
Georgian	100	100	-	100	-	100	100	-	100	-	100	100	-	100	-	100	100	-	100	-
German	89	80	74	67	55	74	57	50	38	21	94	84	71	66	46	100	99	100	98	99
Greek	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100
Gujarati	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100
Hebrew	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100
Hindi	73	33	80	58	51	61	11	65	28	16	64	20	75	49	38	93	67	99	99	98
Hungarian	95	90	88	78	77	87	77	75	53	51	98	94	91	82	82	100	100	100	100	99
Icelandic	93	88	90	76	78	83	72	76	53	53	97	92	94	76	82	100	99	100	99	99
Indonesian	60	48	60	29	18	39	25	37	10	0	61	46	62	25	1	81	72	82	52	54
Irish	91	85	90	78	80	82	70	80	56	58	94	90	92	82	85	96	95	99	97	98
Italian	87	71	80	64	51	69	42	58	31	12	92	74	84	61	43	100	98	99	100	98
Japanese	100	100	25	95	98	100	100	1	87	99	100	100	5	100	100	100	100	68	100	96
Kazakh	92	90	-	85	-	80	78	-	66	-	96	93	-	90	-	99	99	-	100	-
Korean	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100
Latin	87	73	-	70	-	72	49	-	43	-	93	76	-	71	-	97	93	-	96	-
Latvian	93	87	90	86	78	85	75	78	72	56	97	90	93	88	82	99	97	98	98	97
Lithuanian	95	87	89	79	72	86	76	74	56	40	98	89	92	83	77	100	98	99	99	98
Macedonian	84	72	83	68	46	66	52	66	37	10	86	70	83	68	32	99	95	100	98	97
Malay	31	31	23	19	4	26	22	19	10	0	38	36	22	20	0	30	36	28	27	11
Maori	92	83	-	92	-	84	64	-	85	-	92	88	-	90	-	99	98	-	100	-
Marathi	85	41	90	81	71	74	20	81	62	43	85	30	92	83	74	96	72	98	98	96
Mongolian	97	96	-	84	-	93	89	-	66	-	99	98	-	88	-	99	99	-	99	-
Nynorsk	66	52	-	55	-	41	25	-	24	-	66	49	-	47	-	90	81	-	92	-
Persian	90	80	81	75	62	78	62	65	53	29	94	80	79	74	58	100	98	99	99	99
Polish	95	90	90	83	81	85	77	76	61	57	98	93	93	89	86	100	99	100	100	100
Portuguese	81	69	63	58	40	59	42	34	22	7	85	70	58	54	19	98	95	98	98	94
Punjabi	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100
Romanian	87	72	78	67	55	69	49	57	34	24	92	74	80	68	50	99	94	97	99	91
Russian	90	78	80	50	53	76	59	62	20	22	95	84	85	43	50	98	92	94	86	87
Serbian	88	78	73	73	46	74	62	57	46	18	90	80	70	74	39	99	91	90	98	80
Shona	91	81	-	-	-	78	56	-	-	-	96	86	-	-	-	100	100	-	-	-
Slovak	84	75	76	70	47	64	49	53	39	12	90	78	76	73	38	99	97	98	99	92
Slovene	82	67	74	71	37	61	39	53	43	3	87	68	72	72	18	99	93	98	99	90
Somali	92	85	91	69	79	82	64	78	35	50	96	90	94	74	88	100	100	100	98	100
Sotho	85	72	-	-	-	67	43	-	-	-	90	75	-	-	-	99	97	-	-	-
Spanish	70	56	59	42	32	44	26	29	8	0	69	49	50	25	6	97	94	97	93	91
Swahili	81	70	75	73	60	60	43	50	45	26	84	68	75	74	58	98	97	99	99	98
Swedish	84	72	71	69	50	64	46	44	41	15	88	76	72	69	42	99	95	97	97	94
Tagalog	78	66	77	61	61	52	36	53	27	23	83	67	79	57	62	99	96	99	98	97
Tamil	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100
Telugu	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100	100
Thai	99	99	100	100	100	100	100	100	100	100	100	100	100	100	100	98	98	100	99	100
Tsonga	84	72	-	-	-	66	46	-	-	-	89	73	-	-	-	98	97	-	-	-
Tswana	84	71	-	-	-	65	44	-	-	-	88	73	-	-	-	99	96	-	-	-
Turkish	94	87	81	72	70	84	71	62	48	43	98	91	83	71	70	100	99	99	98	96
Ukrainian	92	86	81	79	68	84	75	62	54	39	97	92	84	83	69	95	93	97	99	94
Urdu	91	80	83	68	72	80	65	68	45	49	94	78	84	62	71	98	96	96	98	96
Vietnamese	91	87	85	84	87	79	76	63	66	65	94	87	92	86	95	99	98	100	100	100
Welsh	91	82	85	77	77	78	61	68	50	50	96	87	88	81	82	99	99	100	99	99
Xhosa	82	69	-	-	-	64	45	-	-	-	85	67	-	-	-	98	94	-	-	-
Yoruba	75	62	-	-	-	50	33	-	-	-	77	61	-	-	-	97	93	-	-	-
Zulu	81	70	-	78	-	62	45	-	51	-	83	72	-	82	-	97	94	-	100	-
Mean	86	77	80	74	65	74	61	64	53	41	89	78	81	74	61	96	93	96	95	93
Median	89.23	79.63	81.3	75.55	63.85	74.3	56.7	63.39	48.7	30.75	93.7	80.6	84.25	75.55	65.95	99.0	96.9	99.15	99.0	97.4
Standard Deviation	13.12	17.26	16.2	18.56	23.87	18.43	24.81	23.9	27.37	33.87	13.13	18.96	18.74	21.32	31.32	11.02	11.86	10.77	12.59	13.54

Every language detector uses a probabilistic n-gram model trained on the character distribution in some training corpus. Most libraries only use n-grams of size 3 (trigrams) which is satisfactory for detecting the language of longer text fragments consisting of multiple sentences. For short phrases or single words, however, trigrams are not enough. The shorter the input text is, the less n-grams are available. The probabilities estimated from such few n-grams are not reliable. This is why Lingua makes use of n-grams of sizes 1 up to 5 which results in much more accurate prediction of the correct language.

A second important difference is that Lingua does not only use such a statistical model, but also a rule-based engine. This engine first determines the alphabet of the input text and searches for characters which are unique in one or more languages. If exactly one language can be reliably chosen this way, the statistical model is not necessary anymore. In any case, the rule-based engine filters out languages that do not satisfy the conditions of the input text. Only then, in a second step, the probabilistic n-gram model is taken into consideration. This makes sense because loading less language models means less memory consumption and better runtime performance.

In general, it is always a good idea to restrict the set of languages to be considered in the classification process using the respective api methods. If you know beforehand that certain languages are never to occur in an input text, do not let those take part in the classifcation process. The filtering mechanism of the rule-based engine is quite good, however, filtering based on your own knowledge of the input text is always preferable.

If you want to reproduce the accuracy results above, you can generate the test reports yourself for all four classifiers and all languages by doing:

./gradlew accuracyReport

You can also restrict the classifiers and languages to generate reports for by passing arguments to the Gradle task. The following task generates reports for Lingua and the languages English and German only:

./gradlew accuracyReport -Pdetectors=Lingua -Planguages=English,German

By default, only a single CPU core is used for report generation. If you have a multi-core CPU in your machine, you can fork as many processes as you have CPU cores. This speeds up report generation significantly. However, be aware that forking more than one process can consume a lot of RAM. You do it like this:

./gradlew accuracyReport -PcpuCores=2

For each detector and language, a test report file is then written into /accuracy-reports, to be found next to the src directory. As an example, here is the current output of the Lingua German report:

##### GERMAN #####

Legend: 'low accuracy mode | high accuracy mode'

>>> Accuracy on average: 79.80% | 89.23%

>> Detection of 1000 single words (average length: 9 chars)
Accuracy: 56.70% | 73.90%
Erroneously classified as DUTCH: 2.80% | 2.30%, DANISH: 2.20% | 2.10%, ENGLISH: 1.90% | 2.00%, LATIN: 1.90% | 1.90%, BOKMAL: 2.40% | 1.60%, BASQUE: 1.60% | 1.20%, ITALIAN: 1.00% | 1.20%, FRENCH: 1.60% | 1.20%, ESPERANTO: 1.10% | 1.10%, SWEDISH: 3.20% | 1.00%, AFRIKAANS: 1.30% | 0.80%, TSONGA: 1.50% | 0.70%, NYNORSK: 1.40% | 0.60%, PORTUGUESE: 0.50% | 0.60%, YORUBA: 0.40% | 0.60%, SOTHO: 0.70% | 0.50%, FINNISH: 0.80% | 0.50%, WELSH: 1.30% | 0.50%, SPANISH: 1.20% | 0.40%, SWAHILI: 0.60% | 0.40%, TSWANA: 2.20% | 0.40%, POLISH: 0.70% | 0.40%, ESTONIAN: 0.90% | 0.40%, IRISH: 0.50% | 0.40%, TAGALOG: 0.10% | 0.30%, ICELANDIC: 0.30% | 0.30%, BOSNIAN: 0.10% | 0.30%, LITHUANIAN: 0.80% | 0.20%, MAORI: 0.50% | 0.20%, INDONESIAN: 0.40% | 0.20%, ALBANIAN: 0.60% | 0.20%, CATALAN: 0.70% | 0.20%, ZULU: 0.30% | 0.20%, ROMANIAN: 1.20% | 0.20%, CROATIAN: 0.10% | 0.20%, XHOSA: 0.40% | 0.20%, TURKISH: 0.70% | 0.10%, MALAY: 0.50% | 0.10%, LATVIAN: 0.40% | 0.10%, SLOVENE: 0.00% | 0.10%, SLOVAK: 0.30% | 0.10%, SOMALI: 0.00% | 0.10%, HUNGARIAN: 0.40% | 0.00%, SHONA: 0.80% | 0.00%, VIETNAMESE: 0.40% | 0.00%, CZECH: 0.30% | 0.00%, GANDA: 0.20% | 0.00%, AZERBAIJANI: 0.10% | 0.00%

>> Detection of 1000 word pairs (average length: 18 chars)
Accuracy: 83.50% | 94.10%
Erroneously classified as DUTCH: 1.50% | 0.90%, LATIN: 1.00% | 0.80%, ENGLISH: 1.40% | 0.70%, SWEDISH: 1.40% | 0.60%, DANISH: 1.20% | 0.50%, FRENCH: 0.60% | 0.40%, BOKMAL: 1.40% | 0.30%, TAGALOG: 0.10% | 0.20%, IRISH: 0.20% | 0.20%, TURKISH: 0.10% | 0.10%, NYNORSK: 0.90% | 0.10%, TSONGA: 0.40% | 0.10%, ZULU: 0.10% | 0.10%, ESPERANTO: 0.30% | 0.10%, AFRIKAANS: 0.60% | 0.10%, ITALIAN: 0.10% | 0.10%, ESTONIAN: 0.30% | 0.10%, FINNISH: 0.40% | 0.10%, SOMALI: 0.00% | 0.10%, SWAHILI: 0.20% | 0.10%, MAORI: 0.00% | 0.10%, WELSH: 0.10% | 0.10%, LITHUANIAN: 0.40% | 0.00%, INDONESIAN: 0.10% | 0.00%, CATALAN: 0.30% | 0.00%, LATVIAN: 0.20% | 0.00%, XHOSA: 0.30% | 0.00%, SPANISH: 0.50% | 0.00%, MALAY: 0.10% | 0.00%, SLOVAK: 0.10% | 0.00%, BASQUE: 0.40% | 0.00%, YORUBA: 0.20% | 0.00%, TSWANA: 0.30% | 0.00%, SHONA: 0.10% | 0.00%, PORTUGUESE: 0.10% | 0.00%, SOTHO: 0.30% | 0.00%, CZECH: 0.10% | 0.00%, ALBANIAN: 0.40% | 0.00%, AZERBAIJANI: 0.10% | 0.00%, ICELANDIC: 0.10% | 0.00%, SLOVENE: 0.10% | 0.00%

>> Detection of 1000 sentences (average length: 111 chars)
Accuracy: 99.20% | 99.70%
Erroneously classified as DUTCH: 0.00% | 0.20%, LATIN: 0.20% | 0.10%, NYNORSK: 0.10% | 0.00%, SPANISH: 0.10% | 0.00%, DANISH: 0.10% | 0.00%, SOTHO: 0.20% | 0.00%, ZULU: 0.10% | 0.00%

The plots have been created with Python and the libraries Pandas, Matplotlib and Seaborn. If you have a global Python 3 installation and the python3 command available on your command line, you can redraw the plots after modifying the test reports by executing the following Gradle task:

./gradlew drawAccuracyPlots

The detailed statistics table that contains all accuracy values can be written with:

./gradlew writeAccuracyTable

Lingua is hosted on GitHub Packages and Maven Central.

// Groovy syntax
implementation 'com.github.pemistahl:lingua:1.2.1'

// Kotlin syntax
implementation("com.github.pemistahl:lingua:1.2.1")

<dependency>
    <groupId>com.github.pemistahl</groupId>
    <artifactId>lingua</artifactId>
    <version>1.2.1</version>
</dependency>

Lingua uses Gradle to build and requires Java >= 1.8 for that.

git clone https://github.com/pemistahl/lingua.git
cd lingua
./gradlew build

Several jar archives can be created from the project.

1. ./gradlew jar assembles lingua-1.2.1.jar containing the compiled sources only.
2. ./gradlew sourcesJar assembles lingua-1.2.1-sources.jar containing the plain source code.
3. ./gradlew jarWithDependencies assembles lingua-1.2.1-with-dependencies.jar containing the compiled sources and all external dependencies needed at runtime. This jar file can be included in projects without dependency management systems. It can also be used to run Lingua in standalone mode (see below).

Lingua can be used programmatically in your own code or in standalone mode.

The API is pretty straightforward and can be used in both Kotlin and Java code.

/* Kotlin */

import com.github.pemistahl.lingua.api.*
import com.github.pemistahl.lingua.api.Language.*

val detector: LanguageDetector = LanguageDetectorBuilder.fromLanguages(ENGLISH, FRENCH, GERMAN, SPANISH).build()
val detectedLanguage: Language = detector.detectLanguageOf(text = "languages are awesome")

The public API of Lingua never returns null somewhere, so it is safe to be used from within Java code as well.

/* Java */

import com.github.pemistahl.lingua.api.*;
import static com.github.pemistahl.lingua.api.Language.*;

final LanguageDetector detector = LanguageDetectorBuilder.fromLanguages(ENGLISH, FRENCH, GERMAN, SPANISH).build();
final Language detectedLanguage = detector.detectLanguageOf("languages are awesome");

By default, Lingua returns the most likely language for a given input text. However, there are certain words that are spelled the same in more than one language. The word prologue, for instance, is both a valid English and French word. Lingua would output either English or French which might be wrong in the given context. For cases like that, it is possible to specify a minimum relative distance that the logarithmized and summed up probabilities for each possible language have to satisfy. It can be stated in the following way:

val detector = LanguageDetectorBuilder
    .fromAllLanguages()
    .withMinimumRelativeDistance(0.25) // minimum: 0.00 maximum: 0.99 default: 0.00
    .build()

Be aware that the distance between the language probabilities is dependent on the length of the input text. The longer the input text, the larger the distance between the languages. So if you want to classify very short text phrases, do not set the minimum relative distance too high. Otherwise you will get most results returned as Language.UNKNOWN which is the return value for cases where language detection is not reliably possible.

Knowing about the most likely language is nice but how reliable is the computed likelihood? And how less likely are the other examined languages in comparison to the most likely one? These questions can be answered as well:

val detector = LanguageDetectorBuilder.fromLanguages(GERMAN, ENGLISH, FRENCH, SPANISH).build()
val confidenceValues = detector.computeLanguageConfidenceValues(text = "Coding is fun.")

// {
//   ENGLISH=1.0, 
//   GERMAN=0.8665738136456169, 
//   FRENCH=0.8249537317466078, 
//   SPANISH=0.7792362923625288
// }

In the example above, a map of all possible languages is returned, sorted by their confidence value in descending order. The values that the detector computes are part of a relative confidence metric, not of an absolute one. Each value is a number between 0.0 and 1.0. The most likely language is always returned with value 1.0. All other languages get values assigned which are lower than 1.0, denoting how less likely those languages are in comparison to the most likely language.

The map returned by this method does not necessarily contain all languages which the calling instance of LanguageDetector was built from. If the rule-based engine decides that a specific language is truly impossible, then it will not be part of the returned map. Likewise, if no ngram probabilities can be found within the detector's languages for the given input text, the returned map will be empty. The confidence value for each language not being part of the returned map is assumed to be 0.0.

By default, Lingua uses lazy-loading to load only those language models on demand which are considered relevant by the rule-based filter engine. For web services, for instance, it is rather beneficial to preload all language models into memory to avoid unexpected latency while waiting for the service response. If you want to enable the eager-loading mode, you can do it like this:

LanguageDetectorBuilder.fromAllLanguages().withPreloadedLanguageModels().build()

Multiple instances of LanguageDetector share the same language models in memory which are accessed asynchronously by the instances.

Lingua's high detection accuracy comes at the cost of being noticeably slower than other language detectors. The large language models also consume significant amounts of memory. These requirements might not be feasible for systems running low on resources. If you want to classify mostly long texts or need to save resources, you can enable a low accuracy mode that loads only a small subset of the language models into memory:

LanguageDetectorBuilder.fromAllLanguages().withLowAccuracyMode().build()

The downside of this approach is that detection accuracy for short texts consisting of less than 120 characters will drop significantly. However, detection accuracy for texts which are longer than 120 characters will remain mostly unaffected.

An alternative for a smaller memory footprint and faster performance is to reduce the set of languages when building the language detector. In most cases, it is not advisable to build the detector from all supported languages. When you have knowledge about the texts you want to classify you can almost always rule out certain languages as impossible or unlikely to occur.

There might be classification tasks where you know beforehand that your language data is definitely not written in Latin, for instance (what a surprise :-). The detection accuracy can become better in such cases if you exclude certain languages from the decision process or just explicitly include relevant languages:

// include all languages available in the library
// WARNING: in the worst case this produces high memory 
//          consumption of approximately 3.5GB 
//          and slow runtime performance
//          (in high accuracy mode)
LanguageDetectorBuilder.fromAllLanguages()

// include only languages that are not yet extinct (= currently excludes Latin)
LanguageDetectorBuilder.fromAllSpokenLanguages()

// include only languages written with Cyrillic script
LanguageDetectorBuilder.fromAllLanguagesWithCyrillicScript()

// exclude only the Spanish language from the decision algorithm
LanguageDetectorBuilder.fromAllLanguagesWithout(Language.SPANISH)

// only decide between English and German
LanguageDetectorBuilder.fromLanguages(Language.ENGLISH, Language.GERMAN)

// select languages by ISO 639-1 code
LanguageDetectorBuilder.fromIsoCodes639_1(IsoCode639_1.EN, IsoCode639_3.DE)

// select languages by ISO 639-3 code
LanguageDetectorBuilder.fromIsoCodes639_3(IsoCode639_3.ENG, IsoCode639_3.DEU)

Internally, Lingua efficiently uses all cores of your CPU in order to speed up loading the language models and language detection itself. For this purpose, an internal ForkJoinPool is used. If the library is used within an application server, the consumed memory will not be freed automatically when the application is undeployed.

If you want to free all of Lingua's resources, you will have to do this manually by calling detector.unloadLanguageModels() during the undeployment. This will clear all loaded language models from memory but the thread pool will keep running.

If you want to try out Lingua before you decide whether to use it or not, you can run it in a REPL and immediately see its detection results.

1. With Gradle: ./gradlew runLinguaOnConsole --console=plain
2. Without Gradle: java -jar lingua-1.2.1-with-dependencies.jar

Then just play around:

This is Lingua.
Select the language models to load.

1: enter language iso codes manually
2: all supported languages

Type a number and press <Enter>.
Type :quit to exit.

> 1
List some language iso 639-1 codes separated by spaces and press <Enter>.
Type :quit to exit.

> en fr de es
Loading language models...
Done. 4 language models loaded lazily.

Type some text and press <Enter> to detect its language.
Type :quit to exit.

> languages
ENGLISH
> Sprachen
GERMAN
> langues
FRENCH
> :quit
Bye! Ciao! Tschüss! Salut!

In case you want to contribute something to Lingua, then you are encouraged to do so. Do you have ideas for improving the API? Are there some specific languages that you want to have supported early? Or have you found any bugs so far? Feel free to open an issue or send a pull request. It's very much appreciated.

For pull requests, please make sure that all unit tests pass and that the code is formatted according to the official Kotlin style guide. You can check this by running the Kotlin linter ktlint using ./gradlew ktlintCheck. Most issues which the linter identifies can be fixed by running ./gradlew ktlintFormat. All other issues, especially lines which are longer than 120 characters, cannot be fixed automatically. In this case, please format the respective lines by hand. You will notice that the build will fail if the formatting is not correct.

All kinds of pull requests are welcome. The most favorite ones are new language additions. If you want to contribute new languages to Lingua, here comes a detailed manual explaining how to accomplish that.

Thank you very much in advance for all contributions, however small they may be.

In order to execute the steps below, you will need Java 8 or greater. Even though the library itself runs on Java >= 6, the FilesWriter classes make use of the java.nio api which was introduced with Java 8.

1. Clone Lingua's repository to your own computer as described in section 8.
2. Open enums IsoCode639_1 and IsoCode639_3 and add the language's iso codes. Among other sites, Wikipedia provides a comprehensive list.
3. Open enum Language and add a new entry for your language. If the language is written with a script that is not yet supported by Lingua's Alphabet enum, then add a new entry for it there as well.
4. If your language's script contains characters that are completely unique to it, then add them to the respective entry in the Language enum. However, if the characters occur in more than one language but not in all languages, then add them to the CHARS_TO_LANGUAGES_MAPPING constant in class Constant instead.
5. Use LanguageModelFilesWriter to create the language model files. The training data file used for ngram probability estimation is not required to have a specific format other than to be a valid txt file.
6. Create a new subdirectory in /src/main/resources/language-models and put the generated language model files in there. Do not rename the language model files. The name of the subdirectory must be the language's ISO 639-1 code, completely lowercased.
7. Use TestDataFilesWriter to create the test data files used for accuracy report generation. The input file from which to create the test data should have each sentence on a separate line.
8. Put the generated test data files in /src/accuracyReport/resources/language-testdata. Do not rename the test data files.
9. For accuracy report generation, create an abstract base class for the main logic in /src/accuracyReport/kotlin/com/github/pemistahl/lingua/report/config. Look at the other languages' files in this directory to see how the class must look like. It should be pretty self-explanatory.
10. Create a concrete test class in /src/accuracyReport/kotlin/com/github/pemistahl/lingua/report/lingua. Look at the other languages' files in this directory to see how the class must look like. It should be pretty self-explanatory. If one of the other language detector libraries supports your language already, you can add test classes for those as well. Each library has its own directory for this purpose.
11. Fix the existing unit tests by adding your new language.
12. Add your new language to property linguaSupportedLanguages in /gradle.properties.
13. Run ./gradlew writeAccuracyReports and add the updated accuracy reports to your pull request.
14. Run ./gradlew drawAccuracyPlots and add the updated plots to your pull request.
15. Run ./gradlew writeAccuracyTable and add the updated accuracy table to your pull request.
16. Be happy! :-) You have successfully contributed a new language and have thereby significantly widened this library's fields of application.

Take a look at the planned issues.