Generating Sentences from a Continuous Space

نویسندگان

  • Samuel R. Bowman
  • Luke Vilnis
  • Oriol Vinyals
  • Andrew M. Dai
  • Rafal Józefowicz
  • Samy Bengio
چکیده

The standard recurrent neural network language model (rnnlm) generates sentences one word at a time and does not work from an explicit global sentence representation. In this work, we introduce and study an rnn-based variational autoencoder generative model that incorporates distributed latent representations of entire sentences. This factorization allows it to explicitly model holistic properties of sentences such as style, topic, and high-level syntactic features. Samples from the prior over these sentence representations remarkably produce diverse and well-formed sentences through simple deterministic decoding. By examining paths through this latent space, we are able to generate coherent novel sentences that interpolate between known sentences. We present techniques for solving the difficult learning problem presented by this model, demonstrate its effectiveness in imputing missing words, explore many interesting properties of the model’s latent sentence space, and present negative results on the use of the model in language modeling.

برای دانلود رایگان 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

Linear Formulas in Continuous Logic

We prove that continuous sentences preserved by the ultramean construction (a generalization of the ultraproduct construction) are exactly those sentences which are approximated by linear sentences. Continuous sentences preserved by linear elementary equivalence are exactly those sentences which are approximated in the Riesz space generated by linear sentences. Also, characterizations for linea...

متن کامل

A New Method for Generating Continuous Bivariate Distribution Families

Recently, it has been observed that a new method for generating continuous distributions, T - X family, can be quite effectively used to analyze the data in one dimension. The aim of this study is to generalize this method to two dimensional space so that the marginals would have T - X distributions. So, several examples and properties of this family have been presented. As ...

متن کامل

Fuzzy almost generalized $e$-continuous mappings

In this paper, we introduce and characterize the concept of fuzzy almost generalized $e$-continuous mappings. Several interesting properties of these mappings are also given. Examples and counter examples are also given to illustrate the concepts introduced in the paper. We also introduce the concept of fuzzy $f T_{frac{1}{2}}e$-space, fuzzy $ge$-space, fuzzy regular $ge$-space and  fuzzy gener...

متن کامل

Generating an Indoor space routing graph using semantic-geometric method

The development of indoor Location-Based Services faces various challenges that one of which is the method of generating indoor routing graph. Due to the weaknesses of purely geometric methods for generating indoor routing graphs, a semantic-geometric method is proposed to cover the existing gaps in combining the semantic and geometric methods in this study. The proposed method uses the CityGML...

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

عنوان ژورنال:

دوره   شماره 

صفحات  -

تاریخ انتشار 2016