The Multilingual Sign Language Wordnet

Download the data

The data is available for download in several formats. All files are encoded in UTF-8.

Are you a data scientist working with sign languages for the first time? Please take the time to familiarise yourself with the topic. There are many pitfalls to be aware of. We suggest reading the paper Fox, N., Woll, B. & Cormier, K. Best practices for sign language technology research. Univ Access Inf Soc (2023).

CSV

Note that some fields might contain commas, and are surrounded by quote chars where needed.

• sign_data.csv: Data on the signs themselves. Contains their ID and links toward their video and/or online page in their original resource. Some links point to restricted-access resources which may appear as broken links unless the user is logged in to that resource.
• custom_synsets.csv: Data on newly created synsets that are not part of OMW or GermaNet. This includes synsets for closed-class sign/word types traditionally not covered by wordnets, such as pronouns. Each row represents one synset.
• synset_links_accepted.csv: Each row is a link between one sign and one synset, where the synset is one possible meaning of this sign. The synset ID has a prefix to identify it as either from OMW, Germanet (prefix gm), or a custom synset created for the sign wordnet (prefix ea). The third column gives the confidence of this link.

The possible values for the “confidence” column are as follow:

• manual accept: A human expert has validated this synset link as correct
• manual weak accept: A human expert has marked this synset link as probably correct, but needing cross-validation
• auto accept: This synset link has been validated as correct automatically

TAB files for NLTK

This format allows you to easily import the new languages into the NLTK Wordnet library, and allows you to use NLTK's usual functions with the new languages. NLTK does not allow custom synsets, so this format is missing the synsets we created and the signs linked to them. As per NLTK's requirements, languages are identified by their ISO 639-3 language codes, so, for example, German Sign Language is referred to by the code gsg, rather than its common acronym DGS.

For each language we provide three different formats for textually representing signs as wordnet lemmas: By the type ID used in our resource, by their gloss/keywords, or by their video URL. Note that the video URL format only contains entries for which such a URL is available.

Lang.	ISO	Lemma is Type ID	Lemma is Gloss/keyword	Lemma is Video URL
BSL	bfi	sign_wordnet_video_bfi.tab	sign_wordnet_gloss_bfi.tab	sign_wordnet_video_bfi.tab
DGS	gsg	sign_wordnet_video_gsg.tab	sign_wordnet_gloss_gsg.tab	sign_wordnet_video_gsg.tab
DSGS	sgg	sign_wordnet_video_sgg.tab	sign_wordnet_gloss_sgg.tab	sign_wordnet_video_sgg.tab
GSL	gss	sign_wordnet_video_gss.tab	sign_wordnet_gloss_gss.tab	sign_wordnet_video_gss.tab
LSF	fsl	sign_wordnet_video_fsl.tab	sign_wordnet_gloss_fsl.tab	sign_wordnet_video_fsl.tab
NGT	dse	sign_wordnet_video_dse.tab	sign_wordnet_gloss_dse.tab	sign_wordnet_video_dse.tab
PJM	pso	sign_wordnet_video_pso.tab	sign_wordnet_gloss_pso.tab	sign_wordnet_video_pso.tab
STS	swl	sign_wordnet_video_swl.tab	sign_wordnet_gloss_swl.tab	sign_wordnet_video_swl.tab

Usage examples

Using the NGT sign videos in NLTK:

from nltk.corpus import wordnet as wn
with open("sign_wordnet_video_dse.tab",mode="r",encoding="utf-8") as f:
  wn.custom_lemmas(f,"dse")
wn.synset_from_pos_and_offset("n",2129165).lemma_names("dse")
>>>> ['https://signbank.cls.ru.nl/dictionary/protected_media/glossvideo/NGT/LE/LEEUW-B-22.mp4', 'https://signbank.cls.ru.nl/dictionary/protected_media/glossvideo/NGT/LE/LEEUW-A-1759.mp4']

Using the NGT sign glosses in NLTK:

from nltk.corpus import wordnet as wn
with open("sign_wordnet_gloss_dse.tab",mode="r",encoding="utf-8") as f:
  wn.custom_lemmas(f,"dse")
wn.synset_from_pos_and_offset("n",2129165).lemma_names("dse")
>>>> ['LEEUW-B', 'LION-B', 'LEEUW-A', 'LION-A']
sense=wn.synsets("LION-B",lang="dse")[0]
sense.offset()
>>>> 2129165
sense.pos()
>>>> n
sense.definition()
>>>> large gregarious predatory feline of Africa and India having a tawny coat with a shaggy mane in the male