Feature Fusion Transformer Network for Natural Language Inference

Lei, Sun; Yan, Hengxin

doi:10.1109/icma54519.2022.9856400

Cited by 1 publication

(1 citation statement)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Concatenate matching solves the multiple-premise entailment (MPE) task by transforming it into a singlepremise task by concatenating multiple premises into a single premise. Recently, an inference network model is proposed in [27], that is based on the Transformer model structure. The model retains the nonlocal feature extraction advantages of the self-attention mechanism and combines it with the convolution method to enhance the feature attention of the local domain, and effectively combine local and non-local features.…”

Section: Related Workmentioning

confidence: 99%

Investigating the Use of Different Recurrent Neural Networks for Natural Language Inference in Arabic

Ben-Sghaier¹,

Bakari²,

Néji³

2023

Preprint

View full text Add to dashboard Cite

Natural language inference (NLI) is a subfield of natural language processing (NLP) that involves determining the logical relationship between two pieces of text, usually a premise and a hypothesis. The goal of an NLI system is to classify the inference relationship between the premise and the hypothesis into one of three categories, namely, entailment, contradiction and neutral. An understanding of this relationship is useful for performing several NLP tasks, such as, summarization, question-answering, information retrieval, etc. Given the potential of neural networks to handle complex natural language tasks, and the absence of prior research on their application to NLI in Arabic. In this context, we propose the investigation of different neural network models for the NLI task in Arabic language. Particularly, we carried out experiments using various types of recurrent neural networks, such as, Simple RNN, LSTM, GRU, Bi-LSTM, and BiGRU to find the one that performs best on the NLI task in Arabic. We performed our experiments using the existing datasets for Arabic NLI namely, ArbTE, XNLI, and ArNLI.

show abstract

Section: Related Workmentioning

confidence: 99%