Pat-in-the-Loop: Declarative Knowledge for Controlling Neural Networks

Abstract

The dazzling success of neural networks over natural language processing systems is imposing an urgent need to control their behavior with simpler, more direct declarative rules. In this paper, we propose Pat-in-the-Loop as a model to control a specific class of syntax-oriented neural networks by adding declarative rules. In Pat-in-the-Loop, distributed tree encoders allow to exploit parse trees in neural networks, heat parse trees visualize activation of parse trees, and parse subtrees are used as declarative rules in the neural network. Hence, Pat-in-the-Loop is a model to include human control in specific natural language processing (NLP)-neural network (NN) systems that exploit syntactic information, which we will generically call Pat. A pilot study on question classification showed that declarative rules representing human knowledge, injected by Pat, can be effectively used in these neural networks to ensure correctness, relevance, and cost-effective.

Dis-Cover AI Minds to Preserve Human Knowledge

“…In order to try this test, we need to build machines that are partially innate and therefore, able to learn from experience, as a machine learning paradigm, but at the same time act by human hand. This architecture should be similar to Pat-in-the-Loop [63], a system that allows humans to input rules into a neural network. The results must be interpretable, so humans must be able to understand the decisions made by the system.…”

Exploring Linguistic Properties of Monolingual BERTs with Typological Classification among Languages

“…The rationale behind this trend is that deep architecture training allows rules and structural information about language to emerge directly from sentences in the target language, sacrificing the interpretable and transparent definition of language regularities. Some exceptions exist where structural syntactic information is explicitly encoded in multilayer perceptrons (Zanzotto et al, 2020) with relevant results on unseen sentences (Onorati et al, 2023). Yet, pre-trained transformers (Vaswani et al, 2017;Devlin et al, 2019) are offered as versatile universal sentence/text encoders that contain whatever is needed to solve any downstream task.…”