XLEnt: Mining a Large Cross-lingual Entity Dataset with Lexical-Semantic-Phonetic Word Alignment

This corpus was created by mining CCAligned, CCMatrix, and WikiMatrix parallel sentences. These three sources were themselves extracted from web data from Commoncrawl Snapshots and Wikipedia snapshots. Entity pairs were obtained by performing named entity recognition and typing on English sentences and projecting labels to non-English aligned sentence pairs. No claims of intellectual property are made on the work of preparation of the corpus.

Summary

XLEnt consists of parallel entity mentions in 120 languages aligned with English. These entity pairs were constructed by performing named entity recognition (NER) and typing on English sentences from mined sentence pairs. These extracted English entity labels and types were projected to the non-English sentences through word alignment. Word alignment was performed by combining three alignment signals ((1) word co-occurence alignment with FastAlign (2) semantic alignment using LASER embeddings, and (3) phonetic alignment via transliteration) into a unified word-alignment model. This lexical/semantic/phonetic alignment approach yielded more than 160 million aligned entity pairs in 120 languages paired with English. Recognizing that each English is often aligned to mulitple entities in different target languages, we can join on English entities to obtain aligned entity pairs that directly pair two non-English entities (e.g., Arabic-French)

Citation

If you use the dataset or code, please cite (pdf):

@inproceedings{elkishky_xlent_2021,
 author = {El-Kishky, Ahmed and Renduchintala, Adi and Cross, James and Guzm{\'a}n, Francisco and Koehn, Philipp},
 booktitle = {Preprint},
 title = {{XLEnt}: Mining Cross-lingual Entities with Lexical-Semantic-Phonetic Word Alignment},
 year = {2021}
 address = "Online",
}

Data Format

The entity data is organized into tab separated files from English-to-Target where each row of data is formatted as follows:

Entity_Type \tab Source_Entity \tab Target_Entity \tab Frequency

To obtain aligned entity pairs from non-English to non-English languages, one should simply join two English-aligned entity pair lists on the English entity (Source_Entity).

The annotated sentence pairs use the standard BIO annotation as illustrated in the following example:

I will continue to work as Tony Blair did very closely with the American administration . O O O O O O B-PERSON I-PERSON O O O O O B-NORP O O

Continueriò a lavorare come Tony Blair ha fatto da vicino con l' amministrazione americana . O O O O B-PERSON I-PERSON O O O O O O O B-NORP O

Entity Pairs by Language

All entitity pairs (6.2GB)

en-af (3.8M)
en-am (1.4M)
en-an (464K)
en-ar (134M)
en-arz (444K)
en-as (72K)
en-ast (14M)
en-az (5.7M)
en-ba (692K)
en-bar (408K)
en-be (11M)
en-bg (55M)
en-bn (37M)
en-br (2.0M)
en-bs (4.6M)
en-ca (53M)
en-cb (88K)
en-ceb (3.0M)
en-cs (69M)
en-cx (852K)
en-cy (3.8M)
en-da (50M)
en-de (72M)
en-el (62M)
en-eo (41M)
en-es (171M)
en-et (31M)
en-eu (13M)
en-fa (39M)
en-ff (96K)
en-fi (47M)
en-fo (564K)
en-fr (148M)
en-fy (4.4M)
en-ga (2.6M)
en-gd (932K)
en-gl (30M)
en-gu (644K)
en-ha (6.5M)
en-he (61M)
en-hi (48M)
en-hr (52M)
en-ht (1.9M)
en-hu (65M)
en-hy (4.7M)
en-id (73M)
en-ig (1.0M)
en-ilo (1.3M)
en-io (256K)
en-is (15M)
en-it (105M)
en-ja (138M)
en-jv (3.9M)
en-ka (5.9M)
en-kk (3.6M)
en-km (2.1M)
en-kn (680K)
en-ko (57M)
en-la (3.1M)
en-lb (3.8M)
en-lg (16K)
en-lmo (300K)
en-ln (36K)
en-lo (676K)
en-lt (29M)
en-lv (27M)
en-mg (5.1M)
en-mk (40M)
en-ml (18M)
en-mn (1.9M)
en-mr (12M)
en-ms (31M)
en-mwl (228K)
en-my (1.4M)
en-nds (900K)
en-nds_nl (236K)
en-ne (6.6M)
en-nl (106M)
en-no (36M)
en-ns (28K)
en-oc (3.8M)
en-om (20K)
en-or (428K)
en-pa (584K)
en-pl (102M)
en-ps (1.1M)
en-pt (105M)
en-ro (59M)
en-ru (186M)
en-sd (2.7M)
en-sh (3.6M)
en-si (15M)
en-sk (47M)
en-sl (14M)
en-so (1.2M)
en-sq (22M)
en-sr (30M)
en-ss (36K)
en-su (2.3M)
en-sv (61M)
en-sw (15M)
en-ta (12M)
en-te (3.3M)
en-tg (320K)
en-th (38M)
en-tl (17M)
en-tn (68K)
en-tr (70M)
en-tt (728K)
en-ug (108K)
en-uk (81M)
en-ur (15M)
en-vi (4.0K)
en-wo (116K)
en-wuu (844K)
en-xh (13M)
en-yi (2.1M)
en-yo (812K)
en-zh (127M)
en-zu (464K)

Annotated Sentence Pairs by Language

en-af (120MB)
en-am (7.5MB)
en-an (1.4MB)
en-ar (1.3GB)
en-arz (1.2MB)
en-as (237KB)
en-ast (47MB)
en-az (25MB)
en-ba (1.5MB)
en-bar (1.1MB)
en-be (56MB)
en-bg (556MB)
en-bn (219MB)
en-br (5.5MB)
en-bs (22MB)
en-ca (461MB)
en-cb (305KB)
en-ceb (18MB)
en-cs (741MB)
en-cx (4.3MB)

en-cy (20MB)
en-da (532MB)
en-de (554MB)
en-el (515MB)
en-eo (556MB)
en-es (1.4GB)
en-et (344MB)
en-eu (70MB)
en-fa (387MB)
en-ff (241KB)
en-fi (431MB)
en-fo (1.8MB)
en-fr (1.1GB)
en-fy (13MB)
en-ga (6.4MB)
en-gd (3.1MB)
en-gl (231MB)
en-gu (2.8MB)
en-ha (66MB)
en-he (561MB)

en-hi (385MB)
en-hr (549MB)
en-ht (13MB)
en-hu (682MB)
en-hy (24MB)
en-id (654MB)
en-ig (5.2MB)
en-ilo (6.1MB)
en-io (803KB)
en-is (121MB)
en-it (802MB)
en-ja (940MB)
en-jv (12MB)
en-ka (31MB)
en-kk (15MB)
en-km (8.3MB)
en-kn (2.6MB)
en-ko (311MB)
en-la (9.3MB)
en-lb (52MB)

en-lg (41KB)
en-lmo (751KB)
en-ln (85KB)
en-lo (2.5MB)
en-lt (340MB)
en-lv (141MB)
en-mg (36MB)
en-mk (338MB)
en-ml (55MB)
en-mn (8.1MB)
en-mr (37MB)
en-ms (284MB)
en-mwl (786KB)
en-my (6.0MB)
en-nds (2.7MB)
en-nds_nl (707KB)
en-ne (20MB)
en-nl (1.1GB)
en-no (326MB)
en-ns (105KB)

en-oc (13MB)
en-om (59KB)
en-or (1.1MB)
en-pa (2.7MB)
en-pl (978MB)
en-ps (4.8MB)
en-pt (737MB)
en-ro (722MB)
en-ru (1.4GB)
en-sd (12MB)
en-sh (16MB)
en-si (58MB)
en-sk (562MB)
en-sl (118MB)
en-so (5.3MB)
en-sq (301MB)
en-sr (254MB)
en-ss (116KB)
en-su (12MB)
en-sv (669MB)

en-sw (132MB)
en-ta (76MB)
en-te (16MB)
en-tg (715KB)
en-th (171MB)
en-tl (112MB)
en-tn (189KB)
en-tr (548MB)
en-tt (1.5MB)
en-ug (331KB)
en-uk (746MB)
en-ur (92MB)
en-ur (92MB)
en-wo (334KB)
en-wuu (1.1MB)
en-xh (50MB)
en-yi (22MB)
en-yo (5.1MB)
en-zhX (1.4GB)
en-zu (2.6MB)