Analogical Reasoning: An Algorithm Comparison for Natural Language Processing

Abstract

There is a continual push to make Artificial Intelligence (AI) as human-like as possible; however, this is a difficult task. A significant limitation is the inability of AI to learn beyond its current comprehension. Analogical reasoning (AR), whereby learning by analogy occurs, has been proposed as one method to achieve this goal. Current AR models have their roots in symbolist, connectionist, or hybrid approaches which indicate how analogies are evaluated. No current studies have compared psychologically-inspired and natural language processing (NLP)-produced algorithms to one another; this study compares seven AR algorithms from both realms on multiple-choice word-based analogy problems. Assessment is based on selection of the correct answer, "correctness," and their similarity score prediction compared to the "ideal" score, which is defined as the "goodness" metric. Psychologically-based models have an advantage based on our metrics; however, there is not a clear one-size-fits-all algorithm for all AR problems.

Uncertainty in Visual Generative AI

“…The modified Sternberg and Nigro dataset is particularly fascinating due to its inclusion of abstract and ambiguous concepts such as "true" and "false". The inability to visually represent these concepts has limited visual analogical reasoning research, which is intended to be expanded through the application of AIGC [114]. However, for this research, the individual words within the analogies were used as inputs to the text-to-image model.…”

“…The text-to-image model selected was Craiyon V3 (formerly known as DALL-E Mini), which uses a transformer and generator to create images from a textual prompt [35][36][37]. Craiyon was selected due to having a free tier (unlike Midjourney and DALL-E 2) and considering its previous success established in the literature [114][115][116]. Internally, Craiyon creates its prompt based on the initial prompt to generate nine images per prompt.…”

Integration of Computer Vision with Analogical Reasoning for Characterizing Unknowns

“…The success of analogical reasoning in solving analogy problems has been proven in both the visual/pictorial (Polya, 1990;Zhang, Gao, Baoxiong, Zhu, & Song-Chun, 2019) and text/verbal space (French, 2002;Rogers, Drozd, & Li, 2017). Considerable emphasis has been on the development of analogical reasoning for text-based analogies with many algorithms developed to address the wide range of text-based analogy problems (Combs, Bihl, Ganapathy, & Staples, 2022). These text-based problems range from novel word problems (e.g., king:queen::man:woman) to mapping sentence elements (e.g., "She is growing like a weed") to drawing parallels between stories (Ichien, Lu, & Holyoak, 2020).…”

“…These text-based problems range from novel word problems (e.g., king:queen::man:woman) to mapping sentence elements (e.g., "She is growing like a weed") to drawing parallels between stories (Ichien, Lu, & Holyoak, 2020). Initially, analogical reasoning started as psychologically-based algorithms (see (Gentner, 1983); (Holyoak & Thagard, 1989); (Hofstadter & Mitchell, 1995)) but recently, with the rise of natural language processing, vector space models and artificial neural network approaches have increased in popularity (Combs, Bihl, Ganapathy, & Staples, 2022). To date, the most prominent vector space models include Word2Vec (Mikolov, Sutskever, Chen, Corrado, & Dean, 2013;Mikolov, Tomas, Yih, & Zweig, 2013), Global Vectors (GloVe) (Pennington, Socher, & Manning, 2014), and fastText (Bojanowski, Grave, Joulin, & Mikolov, 2017).…”

“…Within the artificial neural network scope, models include Learning and Inference with Schemas and Analogies (Hummel & Holyoak, 1997) and Structured Tensor Analogical Reasoning 2 (Wilson, Halford, Gray, & Phillips, 2001). A handful of these algorithms were selected for an apples-to-apples comparison which showed the advantages and disadvantages of each (Combs, Bihl, Ganapathy, & Staples, 2022).…”