Amazon yasindikiza pansi pa layisensi ya CC BY 4.0 "MASSIVE" (Mullingual Amazon SLURP for Slot Filling, Intent Classification, and Virtual-assistant Evaluation) dataset, zitsanzo zamakina ophunzirira makina, ndi zida zophunzitsira mitundu yanu yomwe ingagwiritsidwe ntchito kumvetsetsa zambiri zamalankhulidwe achilengedwe (NLU, Natural Language Understanding). Setiyi ili ndi mawu ofotokozera komanso osankhidwa bwino opitilira miliyoni miliyoni okonzedwa m'zinenero 51.
Zotolera za SLURP, zomwe zidalipo m'Chingerezi, zidagwiritsidwa ntchito ngati mawu opangira MASSIVE seti, yomwe idasinthidwa m'zilankhulo zina 50 pogwiritsa ntchito omasulira akatswiri. Tekinoloje ya Alexa ya chilankhulo chachilengedwe (NLU) imatembenuza mawu kukhala mawu, kenako imagwiritsa ntchito mitundu ingapo ya NLU pamawu omwe amasanthula kukhalapo kwa mawu osakira kuti adziwe tanthauzo la funso la wogwiritsa ntchito.
Chimodzi mwazolinga zopanga ndi kufalitsa setiyi ndikusintha othandizira amawu kuti azitha kusanthula zilankhulo zingapo nthawi imodzi, komanso kulimbikitsa opanga gulu lachitatu kuti apange mapulogalamu ndi ntchito zomwe zimakulitsa luso la othandizira mawu. Pofuna kukopa chidwi cha opanga mapulogalamu, Amazon idayambitsa mpikisano kuti apange mtundu wabwino kwambiri wamtundu uliwonse pogwiritsa ntchito deta yosindikizidwa.
Pakadali pano, othandizira mawu amathandizira zilankhulo zochepa chabe ndipo amagwiritsa ntchito mitundu yophunzirira pamakina yolumikizidwa ndi chilankhulo china. Pulojekiti ya MASSIVE ikufuna kuthetsa kuperewera kumeneku popanga zitsanzo zapadziko lonse lapansi ndi makina ophunzirira makina omwe amatha kugawa ndi kukonza zilankhulo zosiyanasiyana.
Source: opennet.ru