Abstract
As stated by Bank (2014 :73), “determining the number of signs” in a
corpus is a “non-trivial task” given that “signers may deviate from
citation forms (by articulating one-handed signs as two-handed and
vice versa), use their non-dominant hands as a buoy, or articulate two
one-handed signs simultaneously”. We propose a semi-automatized
method to establish sign frequency that specifically addresses these
difficulties. This method is replicable for any annotation dataset that
draws on the principles of ID-glossing (Johnston 2016), has two
independent annotation tiers for the hands and does not segment the
buoys as separate annotations (Crasborn et al. 2015).
The main steps are (1) the extraction of the “annotation overlaps
information” from ELAN to Excel, (2) the enrichment of the Excel file
with (a) a unique tag for each annotation and with (b) information
about the handedness of the signs (one-handed and two-handed
signs), (3) the classification of the annotations in three articulatory
categories (one-handed articulations, two-handed articulations,
complex articulations), (4) the frequency count based on the crossing
between handedness information and articulatory pattern information.
The table below shows the ten most frequent signs that were
extracted from the annotations of the Corpus LSFB (Meurant 2015)
with this method.
It allows (1) to tackle the difficulties pointed by Bank (2014) and thus
reduce the sources of noise to the minimum; (2) to revise prior
information about handedness based on usage data (Johnston 2016);
(3) to avoid manual annotation of one-handed and two-handed
variants (Johnston 2016).
corpus is a “non-trivial task” given that “signers may deviate from
citation forms (by articulating one-handed signs as two-handed and
vice versa), use their non-dominant hands as a buoy, or articulate two
one-handed signs simultaneously”. We propose a semi-automatized
method to establish sign frequency that specifically addresses these
difficulties. This method is replicable for any annotation dataset that
draws on the principles of ID-glossing (Johnston 2016), has two
independent annotation tiers for the hands and does not segment the
buoys as separate annotations (Crasborn et al. 2015).
The main steps are (1) the extraction of the “annotation overlaps
information” from ELAN to Excel, (2) the enrichment of the Excel file
with (a) a unique tag for each annotation and with (b) information
about the handedness of the signs (one-handed and two-handed
signs), (3) the classification of the annotations in three articulatory
categories (one-handed articulations, two-handed articulations,
complex articulations), (4) the frequency count based on the crossing
between handedness information and articulatory pattern information.
The table below shows the ten most frequent signs that were
extracted from the annotations of the Corpus LSFB (Meurant 2015)
with this method.
It allows (1) to tackle the difficulties pointed by Bank (2014) and thus
reduce the sources of noise to the minimum; (2) to revise prior
information about handedness based on usage data (Johnston 2016);
(3) to avoid manual annotation of one-handed and two-handed
variants (Johnston 2016).
| Original language | English |
|---|---|
| Publication status | Published - 30 Jul 2018 |
| Event | First International Workshop on Cognitive And Functional Explorations in Sign Language Linguistics - University of Birmingham, Birmingham, United Kingdom Duration: 30 Jul 2018 → 31 Jul 2018 https://www.birmingham.ac.uk/schools/edacs/departments/englishlanguage/events/2018/sign-cafe.aspx |
Conference
| Conference | First International Workshop on Cognitive And Functional Explorations in Sign Language Linguistics |
|---|---|
| Abbreviated title | Sign CAFÉ 1 |
| Country/Territory | United Kingdom |
| City | Birmingham |
| Period | 30/07/18 → 31/07/18 |
| Internet address |