Extending Machine Language Models toward Human-Level Language Understanding

By James L. McClelland, Felix Hill, Maja R. Rudolph, Jason Baldridge, Hinrich Schütze. 2019

37 citations recorded by the source corpus.

Language is central to human intelligence. We review recent break- throughs in machine language processing and consider what re- mains to be achieved. Recent approaches rely on domain general principles of learning and representation captured in artificial neu- ral networks. Most current models, however, focus too closely on language itself. In humans, language is part of a larger system for acquiring, representing, and communicating about objects and sit- uations in the physical and social world, and future machine lan- guage models should emulate such a system. We describe exist- ing machine models linking language to concrete situations, and point toward extensions to address more abstract cases. Human language processing exploits complementary learning systems, in- cluding a deep neural network-like learning system that learns grad- ually as machine systems do, as well as a fast-learning system that supports learning new information quickly. Adding such a system to machine language models will be an important further step toward truly human-like language understanding.

Subfields: Artificial Intelligence, Large Language Models

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Large Language Models

Citations in the corpus

Experience Grounds Language (2020, background): “toward a science (Steedman, 2008), and its major challenges (Hirschberg and Manning, 2015; McClelland et al., 2019), we hope to provide momentum for a direction many are already heading.”
Imitating Interactive Intelligence (2020, methodology): “Here, we have taken steps to construct a more general programme of AI research that emphasises the holistic integration of perception, goal-directed, embodied control, and natural language processing, as has been advocated for previously (McClelland et al., 2019; Lake and Murphy, 2020).”
Probing Contextual Language Models for Common Ground with Visual Representations (2020, background): “We further investigate the inﬂuence of context by evaluating representations when the objects have none or at least one adjective associated with them, as processed by the dependency parser from Al-lenNLP library (Gardner et al., 2018).”
PIGLeT: Language Grounding Through Neuro-Symbolic Interaction in a 3D World (2021, background): “cent work arguing that no amount of language form could ever specify language meaning (McClelland et al., 2019; Bender and Koller, 2020; Bisk et al., 2020); connecting back to the Symbol Grounding Problem of Harnad (1990).”
Cognitive Architectures for Language Agents (2023, background): “Notably, while several recent papers propose conceptual architectures for general intelligence (LeCun, 2022; McClelland et al., 2019) or empirically survey language models and agents (Mialon et al., 2023; Weng, 2023; Wang et al., 2023b), this paper combines elements of both: we propose a…”
Probing Text Models for Common Ground with Visual Representations (2020, background): “Our senses, and notably vision, are essential in shaping the meaning of many of our words, and as such are central for learning semantics (Harnad, 1990; McClelland et al., 2019; Bisk et al., 2020).”
Word meaning in minds and machines (2020, background): “Reviewing evidence from cognitive neuroscience, McClelland et al. (2019) argue that language representations are deeply integrated with multisensory perceptual representations, as well as representations of situations and events.”
Evaluating Machines by their Real-World Language Use (2020, background): “Others in the community have observed similar issues, raising concerns about what today’s benchmark datasets measure (Yogatama et al., 2019; Kryscinski et al., 2019; McClelland et al., 2019; Gardner et al., 2019).”