US 12,135,951 B2
Interpreting cross-lingual models for natural language inference
Xuchao Zhang, Elkridge, MD (US); Bo Zong, West Windsor, NJ (US); Haifeng Chen, West Windsor, NJ (US); and Yanchi Liu, Monmouth Junction, NJ (US)
Assigned to NEC Corporation, Tokyo (JP)
Filed by NEC Laboratories America, Inc., Princeton, NJ (US)
Filed on Jan. 24, 2022, as Appl. No. 17/582,464.
Claims priority of provisional application 63/143,277, filed on Jan. 29, 2021.
Claims priority of provisional application 63/141,011, filed on Jan. 25, 2021.
Prior Publication US 2022/0237391 A1, Jul. 28, 2022
Int. Cl. G06F 40/58 (2020.01); G06F 40/284 (2020.01); G06F 40/30 (2020.01)
CPC G06F 40/58 (2020.01) [G06F 40/284 (2020.01); G06F 40/30 (2020.01)] 12 Claims
OG exemplary drawing
 
9. A computer system for Cross-lingual Transfer Interpretation (CTI), comprising:
a processor;
a display operatively coupled to the processor;
computer memory operatively coupled to the processor; and
a comparator stored in the computer memory, wherein the comparator is configured to receive text corpus data including premise-hypothesis pairs with a relationship label in a source language;
conduct a source to target language translation;
perform a feature importance extraction, where an integrated gradient is applied to assign an importance score to each input feature;
perform a cross-lingual feature alignment, where tokens in the source language are aligned with tokens in the target language for both the premise and the hypothesis based on semantic similarity, wherein the semantic similarity is measured using the cosine similarity,

OG Complex Work Unit Math
where T is the transpose of the vector, and where esi and etj are embeddings of si and tj in such a shared semantic space; and
perform a qualitative analysis, where the importance score of each token can be compared between the source language and the target language according to a feature alignment result.