Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
119 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Improving Neural Question Generation using Answer Separation (1809.02393v2)

Published 7 Sep 2018 in cs.CL, cs.AI, and cs.NE

Abstract: Neural question generation (NQG) is the task of generating a question from a given passage with deep neural networks. Previous NQG models suffer from a problem that a significant proportion of the generated questions include words in the question target, resulting in the generation of unintended questions. In this paper, we propose answer-separated seq2seq, which better utilizes the information from both the passage and the target answer. By replacing the target answer in the original passage with a special token, our model learns to identify which interrogative word should be used. We also propose a new module termed keyword-net, which helps the model better capture the key information in the target answer and generate an appropriate question. Experimental results demonstrate that our answer separation method significantly reduces the number of improper questions which include answers. Consequently, our model significantly outperforms previous state-of-the-art NQG models.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (4)
  1. Yanghoon Kim (4 papers)
  2. Hwanhee Lee (36 papers)
  3. Joongbo Shin (14 papers)
  4. Kyomin Jung (76 papers)
Citations (166)

Summary

Improving Neural Question Generation using Answer Separation

The paper, "Improving Neural Question Generation using Answer Separation," presents a novel approach to enhance Neural Question Generation (NQG) by separating the answer from the passage in the input sequence. Authored by Yanghoon Kim, Hwanhee Lee, Joongbo Shin, and Kyomin Jung at Seoul National University, the paper addresses critical limitations in earlier NQG models, which often result in questions containing answer words unintentionally. Such issues detract from the objective of generating natural, contextually appropriate questions.

The authors propose the answer-separated seq2seq model, which significantly mitigates the problem of answer words appearing in generated questions. This model substitutes the target answer in the passage with a special token to focus the learning process on the context around the target, thereby guiding the model in generating appropriate interrogative words. A key innovation in their approach is the introduction of a module termed "keyword-net." This module is designed to extract and utilize the key information from the target answer, improving the model's ability to generate relevant questions.

In assessing the model's effectiveness, the authors employ the Stanford Question Answering Dataset (SQuAD) to benchmark their results. The answer-separated seq2seq model achieves substantial gains over previous state-of-the-art models in metrics such as BLEU-4, METEOR, and ROUGE-L. The significance of this improvement is underscored by a detailed empirical analysis showing reduced inclusion of target answers in questions, improved prediction of interrogative words, and concentrated attention on interrogative words during the generation process.

The implications of this research are expansive. Practically, it enhances the utility of automated educational systems by generating more precise and meaningful questions. Theoretically, it offers insight into handling the inherent challenge in machine learning of disentangling multiple intertwined inputs to focus on generating a relevant output. Furthermore, this advancement in the field of NQG has potential applications in enhancing question-answering systems and chatbots which aim for engaging, logically evolving dialogue.

Future research directions include further refinement of answer separation techniques and exploration of keyword-net's integration with additional contextual embeddings to enrich model performance. Additionally, extending this method to other domains in natural language processing where the separation of input sequences can yield improved task-specific performance could be promising. As AI progresses, the insights from this paper could influence a variety of applications requiring nuanced comprehension and generation capabilities.