Teaching Machines to Describe Images via Natural Language Feedback

نویسندگان

  • Huan Ling
  • Sanja Fidler
چکیده

Robots will eventually be part of every household. It is thus critical to enable algorithms to learn from and be guided by non-expert users. In this paper, we bring a human in the loop, and enable a human teacher to give feedback to a learning agent in the form of natural language. We argue that a descriptive sentence can provide a much stronger learning signal than a numeric reward in that it can easily point to where the mistakes are and how to correct them. We focus on the problem of image captioning in which the quality of the output can easily be judged by non-experts. We propose a hierarchical phrase-based captioning model trained with policy gradients, and design a feedback network that provides reward to the learner by conditioning on the human-provided feedback. We show that by exploiting descriptive feedback our model learns to perform better than when given independently written human captions.

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Impact of Direct Corrective Feedback (DCF) Through Electronic Portfolio (EP) Platform on the components of Iranian EFL Learners’ Writing across Levels of Language Proficiency

While some researchers have questioned the efficacy of corrective feedback (CF), other researchers believe that CF can be effective if implemented through new technology types, including e-portfolio (EP). However, whether EP can be used as a medium of providing CF for language learners at different levels of language proficiency is still unknown. The purpose of the present study, therefore, was...

متن کامل

Impact of Prompts as Corrective Feedback Strategy on Teaching /θ/ and /ð/ among Iranian Intermediate EFL Learners

This study investigated the effects of prompts as corrective feedback strategy on teaching /θ/ and /ð/ sounds to Iranian EFL learners. To achieve this objective, after 30 students studying English at a language institute took a placement test, the intermediate-level students were selected based on their scores on this test. They were randomly assigned to one experimental group and one control g...

متن کامل

The Effect of Asynchronous versus Computer-mediated Corrective Feedback on the Correct Use of English Articles in an EFL Context

 The purpose of this study is to investigate the effects of asynchronous computer-mediated versus conventional corrective feedback on learners' writing accuracy. Three groups of learners took part in the study: asynchronous feedback group, conventional feedback group, and a control group. Asynchronous feedback group students received explicit feedback on the targeted structure via e-mail, while...

متن کامل

Persian Speakers’ Recognition of English Relative Clauses: The Effects of Enhanced Input vs. Explicit Feedback Types

Despite consensus in focus on form (FOF) instruction over the facilitative role of noticing, controversy has not quelled over ways of directing EFL learners’ attention towards formal features via implicit techniques like input-enhancement or explicit metacognitive feedback and interactive peer-editing on the output they produce. This quasi-experimental study investigated the impact of input enh...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

عنوان ژورنال:
  • CoRR

دوره abs/1706.00130  شماره 

صفحات  -

تاریخ انتشار 2017