A Federated Approach to Predicting Emojis in Hindi Tweets

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 1251 - 1265 December 7-11, 2022 ©2022 Association for Computational Linguistics Successive Prompting for Decomposing Complex Questions Dheeru Dua♣ Shivanshu Gupta♣ Sameer Singh♣, ♠ Matt Gardner♡ ♣University of California, Irvine, USA ♠Allen Institute for Artificial Intelligence ♡Microsoft Semantic Machines {ddua,shivag5,sameer}@uci.edu, mattgardner@microsoft.com Abstract Answering complex questions that require mak- ing latent decisions is a challenging task, es- pecially when limited supervision is available. Recent works leverage the capabilities of large language models (LMs) to perform complex question answering in a few-shot setting by demonstrating how to output intermediate ra- tionalizations while solving the complex ques- tion in a single pass. We introduce “Succes- sive Prompting”, where we iteratively break down a complex task into a simple task, solve it, and then repeat the process until we get the final solution. Successive prompting decou- ples the supervision for decomposing complex questions from the supervision for answering simple questions, allowing us to (1) have multi- ple opportunities to query in-context examples at each reasoning step (2) learn question de- composition separately from question answer- ing, including using synthetic data, and (3) use bespoke (fine-tuned) components for reason- ing steps where a large LM does not perform well. The intermediate supervision is typically manually written, which can be expensive to collect. We introduce a way to generate a syn- thetic dataset which can be used to bootstrap a model’s ability to decompose and answer inter- mediate questions. Our best model (with suc- cessive prompting) achieves an improvement of ∼5% absolute F1 on a few-shot version of the DROP dataset when compared with a state- of-the-art model with the same supervision. 1 Introduction Compositional reading comprehension datasets like HotpotQA (Yang et al., 2018) and DROP (Dua et al., 2019) have inspired a range of model archi- tectures that learn to answer complex questions with weak supervision from the final answer. One recent direction is to leverage large language mod- els (LMs) to solve compositional tasks with very few examples by generating latent reasoning steps before answering the question (Wei et al., 2022; Who kicked the longest field goal in the first half? Q: What are all the field goals in first half? A: 12-yard, 42-yard and 33-yard Q: What is the largest value in: 12-yard, 42-yard and 33-yard? A: 42-yard Q: Who kicked the 42-yard field goal? A: Matt Bryant There are no more questions left to ask. The final answer is Matt Bryant Figure 1: Example decomposition used by Successive Prompting’s question decomposition and question an- swering stage on a DROP example. The model iterates between predicting a simple question to ask and answer- ing the simple question. Nye et al., 2021; Karpas et al., 2022). Given a complex question, this approach first finds nearest- neighbor training examples from a dataset of (ques- tion, reasoning, answer) triples and then concate- nates them to create an input for the LM. A large LM is then prompted with this input to generate the …

Content truncated to 500 words