Learning Language Games through Interaction

Abstract

We introduce a new language learning setting relevant to building adaptive natural language interfaces. It is inspired by Wittgenstein's language games: a human wishes to accomplish some task (e.g., achieving a certain configuration of blocks), but can only communicate with a computer, who performs the actual actions (e.g., removing all red blocks). The computer initially knows nothing about language and therefore must learn it from scratch through interaction, while the human adapts to the computer's capabilities. We created a game called SHRDLURN in a blocks world and collected interactions from 100 people playing it. First, we analyze the humans' strategies, showing that using compositionality and avoiding synonyms correlates positively with task performance. Second, we compare computer strategies, showing that modeling pragmatics on a semantic parsing model accelerates learning for more strategic players.

The Fast and the Flexible: Training Neural Networks to Learn to Follow Instructions from Small Data

“…"red" becomes "roze"-on an online training regime. Finally, we proceeded to learn from real human utterances using the dataset collected by Wang et al (2016).…”

“…We then experiment with different neural network architectures to find which general learning system adapts best for this task. Then, we propose how to adapt this network by training it online and confirm its effectiveness on recovering the meaning of scrambled words and on learning to process the language from human users, using the dataset introduced by Wang et al (2016).…”

“…Instead, we would like these inductive biases to be learned automatically without human intervention. That is, humans should be free from the burden of Figure 1: Illustration of the SHRDLURN task of Wang et al (2016) thinking what are useful primitives for a given domain, but still obtain systems that can learn fast from little data.…”

Pre-Learning Environment Representations for Data-Efficient Neural Instruction Following

“…For our first test environment, we use the block stacking task introduced by Wang et al (2016) and depicted in Figure 1. This environment consists of a series of levels (tasks), where each level requires adding or removing blocks to get from a start configuration to a goal configuration.…”

“…28.5 Human-Designed Representations Wang et al (2016) 33.8 priate for the online learning setup we used in the previous task, so we instead adopt a slightly different evaluation where accuracy at different data sizes is compared. We structure the data for evaluation as follows: First, we group the data so that each group contains only a small number of instructions (10).…”