The Dark Side of the Language: Syntax-Based Neural Networks Rivaling Transformers in Unseen Sentences

October 2023

Dario Onorati, Leonardo Ranaldi, Aria Nourbakhsh, Arianna Patrizi, Elena Sofia Ruzzetti, Francesca Fallucchi, Francesca Fallucchi, Fabio Massimo Zanzotto

Studysummary In this study, the authors found that syntax-based neural networks performed comparably to pre-trained Transformers on tasks involving definitely unseen sentences, suggesting they are a more transparent and parameter-efficient alternative for certain Natural Language Processing applications.

Automatically generated from the study's abstract, not written by a person, and not a review of the full paper. Not medical advice or a treatment recommendation. Read the original study, and consult a qualified healthcare professional before changing treatment. Full disclaimer ·

View this study on iris.uniroma1.it →

Discuss this study in the Community →

Similar Research

5 / 1000+ results

research The Dark Side of the Language: Pre-trained Transformers in the DarkNet

January 2023

This study suggests that pre-trained Transformers only outperform syntactic and lexical neural networks on unseen DarkNet sentences after extreme domain adaptation, indicating unexpected advantages from their massive pre-training corpora.

research Minimizing Factual Inconsistency and Hallucination in Large Language Models

4 citations , November 2023 in “ArXiv.org”

This study demonstrates that a proposed multi-stage framework improves the accuracy and faithfulness of drug-related responses generated by language models, compared to traditional methods.

Inferring Spatially Resolved Transcriptomics Data from Whole Slide Images for the Assessment of Colorectal Tumor Metastasis: A Feasibility Study

research Inferring Spatially Resolved Transcriptomics Data from Whole Slide Images for the Assessment of Colorectal Tumor Metastasis: A Feasibility Study

November 2022 in “bioRxiv (Cold Spring Harbor Laboratory)”

In this study, deep learning models accurately predicted gene expression in whole slide images of colorectal cancer, with convolutional neural networks outperforming transformer and graph-based approaches in spatial RNA pattern prediction.

research Semantic textual similarity using maximal weighted bipartite graph matching

9 citations , June 2012 in “Joint Conference on Lexical and Computational Semantics”

This study presents a system for evaluating semantic similarity between sentences by using token-based matching and improved similarity measures beyond WordNet.

research Towards Fairer Health Recommendations: finding informative unbiased samples via Word Sense Disambiguation

September 2024 in “arXiv (Cornell University)”

This study evaluated various NLP models for detecting bias in medical curricula, finding that fine-tuned BERT models perform well, whereas LLMs, despite being state-of-the-art in many tasks, are unsuitable for this application.