Negation in Cognitive Reasoning

Schon, Claudia; Siebert, Sophie; Stolzenburg, Frieder

doi:10.1007/978-3-030-87626-5_16

Cited by 4 publications

(2 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Negated Examples. Understanding negation is often considered as the first test case in natural language understanding tasks (Ettinger, 2020;Khemlani et al, 2012;Schon et al, 2021). To examine whether PLMs can handle negation in the deductive reasoning task, we construct a set of negated samples by negating the hypothesis h or the cloze prompt c while keeping the premises P unchanged (Hosseini et al, 2021).…”

Section: Adversarial Probingmentioning

confidence: 99%

Can Pretrained Language Models (Yet) Reason Deductively?

Yuan,

Hu,

Vulić

et al. 2023

Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics

View full text Add to dashboard Cite

Acquiring factual knowledge with Pretrained Language Models (PLMs) has attracted increasing attention, showing promising performance in many knowledge-intensive tasks. Their good performance has led the community to believe that the models do possess a modicum of reasoning competence rather than merely memorising the knowledge. In this paper, we conduct a comprehensive evaluation of the learnable deductive (also known as explicit) reasoning capability of PLMs. Through a series of controlled experiments, we posit two main findings. (i) PLMs inadequately generalise learned logic rules and perform inconsistently against simple adversarial surface form edits. (ii) While the deductive reasoning fine-tuning of PLMs does improve their performance on reasoning over unseen knowledge facts, it results in catastrophically forgetting the previously learnt knowledge. Our main results suggest that PLMs cannot yet perform reliable deductive reasoning, demonstrating the importance of controlled examinations and probing of PLMs' deductive reasoning abilities; we reach beyond (misleading) task performance, revealing that PLMs are still far from robust reasoning capabilities, even for simple deductive tasks.

show abstract

Section: Adversarial Probingmentioning

confidence: 99%

Can Pretrained Language Models (Yet) Reason Deductively?

Yuan,

Hu,

Vulić

et al. 2023

Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics

View full text Add to dashboard Cite

show abstract

“…This is also highlighted through the under-representation of negatives in existing natural language inference benchmarks [17,47] and that pretrained language models have difficulty performing well during neural translation tasks [16] and fill-in-the-blank tests [21]. Understanding negations, though harder for learning-based models [51], is crucial for commonsense reasoning tasks [47,48]. This ability is highly desirable in image-text retrieval and text-to-image generation systems [45] which employ VLMs.…”

Section: Introductionmentioning

confidence: 99%

1047Epidemiology of different Cancers in Rural Jodhpur (India)

Singh¹,

Raghav²

2021

International Journal of Epidemiology

View full text Add to dashboard Cite

Background Cancer is the second leading cause of death in India accounting for nearly 9.6 million deaths in 2018. Crude annual incidence rate of all cancers together in Rajasthan has increased from 1990 to 2016 from 58.8 to 72.6 per lakh population. Few studies done are based on hospitals and pathology centres, but the data is limited. Community based study is required to know the true burden and trend of cancer prevalent in the region. Methods A sampling frame was prepared from all the blocks in rural Jodhpur, one block was selected by simple random sampling. Then another sampling frame was prepared by enlisting the villages in the selected block, one third villages were selected by simple random sampling. Results Based on interim analysis, 25 villages have been surveyed till date, with the population of 57,856. We found 91 (0.16%) cancer patients, out of which burden of Oral cancer (19,20.87%) was highest followed by Breast cancer (17,19.31%), Laryngeal cancer (10,10.98%), Primary brain tumours (6,6.59%) respectively. Other cancers were Oesophageal cancer (5) Colorectal cancer (5) Cervical cancer (4) Leukemia (4) Bone tumours (4) Stomach cancer (4) Intestinal cancer (3) Pharyngeal cancer (3) Urinary bladder cancer (1) Thyroid cancer (1) Uterus cancer (1) Lung cancer (1) Renal cell carcinoma (1) Hepatocellular carcinoma (1) and Skin Cancer (1).Tobacco and Opium consumption were found to be associated with Oral and Laryngeal cancer. Conclusions According to the World Cancer Report 2020, Lung cancer is the leading cause of mortality followed by Breast cancer Colorectal cancer, Prostate cancer, Stomach cancer and Cervical cancer which is different from the pattern of Western India. Key messages This study highlights the common cancers and the need to have a public health approach to identify or create awareness for quick treatment.

show abstract

Different negations and fuzziness in web resources and resource description and their mathematical foundation

Pan,

Wang

2025

Information Sciences

View full text Add to dashboard Cite

Negation in Cognitive Reasoning

Cited by 4 publications

References 16 publications

Can Pretrained Language Models (Yet) Reason Deductively?

Can Pretrained Language Models (Yet) Reason Deductively?

1047Epidemiology of different Cancers in Rural Jodhpur (India)

Different negations and fuzziness in web resources and resource description and their mathematical foundation

Contact Info

Product

Resources

About