Semantic tagging

This page is to keep a list of ideas for semantic tagging in Apertium.

Approaches

• Giellatekno
• Grammatical Framework

Data sources

• WikiData ?
  • https://www.wikidata.org/wiki/Q169
  • https://www.wikidata.org/wiki/Q200266

• Embeddings ?

Implementation

Bilingual or multilingual

• Often a word can be disambiguated using its translation in another language, for example the triple (estació, gare, station) defines a building meaning.

Existing examples

SET MangoFruitWords = ("aguacate"i)  OR ("albahaca"i) OR ("alimentario"i) OR ("alimenticio"i) OR ("aloe"i) OR ("anacardo"i) OR ("ananás"i) OR ("anchoa"i) OR (
"arroz"i) OR ("atún"i) OR ("azúcar"i) OR ("banana"i) OR ("banano"i) OR ("batido"i) OR ("boniato"i) OR ("brocheta"i) OR ("cacahuete"i) OR ("cacao"i) OR ("caram
elizar"i) OR ("caramelo"i) OR ("carpaccio"i) OR ("caviar"i) OR ("cereal"i) OR ("chirimoya"i) OR ("chocolate"i) OR ("<chutney>"i) OR ("clima"i) OR ("coco"i) OR
 ("cocotero"i) OR ("codorniz"i) OR ("comer"i) OR ("comercial"i) OR ("comida"i) OR ("cosecha"i) OR ("crema"i) OR ("cultivar"i) OR ("cultivo"i) OR ("cítrico"i) 
OR ("dátil"i) OR ("deshidratar"i) OR ("ensalada"i) OR ("exportación"i) OR ("foie"i) OR ("fragancia"i) OR ("fresa"i) OR ("fresco"i) OR ("fruta"i) OR ("fruto"i)
 OR ("gamba"i) OR ("gazpacho"i) OR ("guayaba"i) OR ("gustar"i) OR ("helado"i) OR ("hortaliza"i) OR ("ingrediente"i) OR ("jamón"i) OR ("jarabe"i) OR ("jardín"i
) OR ("jengibre"i) OR ("judía"i) OR ("langosta"i) OR ("langostino"i) OR ("lechuga"i) OR ("legumbre"i) OR ("maduro"i) OR ("mandarina"i) OR ("mandioca"i) OR ("m
anzana"i) OR ("maní"i) OR ("maracuyá"i) OR ("maíz"i) OR ("melocotón"i) OR ("melón"i) OR ("mono"i) OR ("naranja"i) OR ("naranjo"i) OR ("orquídea"i) OR ("orégan
o"i) OR ("palma"i) OR ("palmera"i) OR ("papaya"i) OR ("parmesano"i) OR ("patata"i) OR ("piscina"i) OR ("piña"i) OR ("plantación"i) OR ("plátano"i) OR ("pollo"
i) OR ("probar"i)OR ("puré"i) OR ("rodaja"i) OR ("ron"i) OR ("salsa"i) OR ("sorbete"i) OR ("sorgo"i) OR ("subsistencia"i) OR ("sésamo"i) OR ("tabaco"i) OR ("t
empura"i) OR ("tomate"i) OR ("trigo"i) OR ("triturar"i) OR ("tropical"i) OR ("tubérculo"i) OR ("vainilla"i) OR ("vinagre"i) OR ("yogur"i)  OR ("zumo"i) OR ("Á
frica");

SET MangoNotFruitWords = ("acero"i) OR ("azada"i) OR ("levantar"i) OR ("alzar"i) OR ("plata"i) OR ("arpón"i) OR ("azote"i) OR ("cuerno"i) OR ("bastón"i) OR ("
bolsa"i) OR ("brazo"i) OR ("silla"i) OR ("centímetro"i) OR ("cinturón"i) OR ("llave"i) OR ("clavar"i) OR ("cubierto"i) OR ("golpear"i) OR ("cuchillo"i) OR ("c
orazón"i) OR ("cuerda"i) OR ("cuerpo"i) OR ("cocina"i) OR ("cuero"i) OR ("cuchara"i) OR ("corto"i) OR ("hacha"i) OR ("herramienta"i) OR ("emplear"i) OR ("empu
ñar"i) OR ("escoba"i) OR ("espada"i) OR ("estirar"i) OR ("apretar"i) OR ("extremo"i) OR ("trabajo"i) OR ("meter"i) OR ("fuego"i) OR ("forma"i) OR ("látigo"i) 
OR ("hoja"i) OR ("madera"i) OR ("cuchillo"i) OR ("girar"i) OR ("escoba"i) OR ("grabar"i) OR ("grueso"i) OR ("instrumento"i) OR ("marfil"i) OR ("lanza"i) OR ("
lanzar"i) OR ("largo"i) OR ("maza"i) OR ("martillo"i) OR ("metal"i) OR ("mover"i) OR ("movimiento"i) OR ("navaja"i) OR ("limpiar"i) OR ("sartén"i) OR ("paella
"i) OR ("palo"i) OR ("pala"i) OR ("papel"i) OR ("pieza"i) OR ("piedra"i) OR ("pequeño"i) OR ("picar"i) OR ("pistola"i) OR ("plástico"i) OR ("plata"i) OR ("plu
ma"i) OR ("puerta"i) OR ("precioso"i) OR ("punta"i) OR ("puñal"i) OR ("cepillo"i) OR ("cepillo"i) OR ("redondo"i) OR ("ropa"i) OR ("rueda"i) OR ("sujetar"i) O
R ("mesa"i) OR ("atravesar"i) OR ("utilizar"i) OR ("alrededor"i) OR ("marfil"i);

SELECT:mango_fruta ("mango_fruta"i) IF (0 ("mango_fruta"i)) (0*/* MangoFruitWords) (NOT 0* MangoNotFruitWords) ;
REMOVE:mango_0 ("mango_fruta"i) IF (0 ("mango"i)) (0 ("mango_fruta"i)) ;

Ideas and notes

Thanks Xavi for the ideas...

What I've been thinking about is a module that would go after
biltrans and before lexical selection. It would essentially reweight
the possible translations based on a bag of words over a fixed
window of words or "sentences" (delimited with '.').

You could have source and target components, so e.g. you might
say that "fruit" is a semantic field or domain which includes,

"mango", "manzana", "plátano", "naranja", ...

and

"mango", "taronja", "poma"

In Catalan. These would be in the monolingual pairs. The
module would take both lists and the input

^querer<vblex><pri><p3><sg>/voler<vblex><pri><p3><sg>$
^mango<n><m><pl>/mànec<n><m><pl>/mango<n><m><pl>$
^y<cnjcoo>/i<cnjcoo>$
^manzana<n><f><pl>/poma<n><f><pl>$

And try and maximise semantic coherence, then it could reweight,
so e.g.

^querer<vblex><pri><p3><sg>/voler<vblex><pri><p3><sg>$
^mango<n><m><pl>/mango<n><m><pl><2.0>/mànec<n><m><pl><0.0>$
^y<cnjcoo>/i<cnjcoo>$
^manzana<n><f><pl>/poma<n><f><pl>$

And pass it to the lexical selection module which will choose the
one with the highest weight.

This would mean a new module, but it would require only minor
changes to the bilingual dictionary and lexical selection, and
wouldn't have any effect on transfer.