什麼是 Web 語言模型 API?What is the Web Language Model API? (預覽)(Preview)

重要

Web 語言模型預覽已於 2018 年 8 月 9 日解除委任。The Web Language Model preview was decommissioned on August 9, 2018. 我們建議使用 Azure Machine Learning 文字分析模組來進行文字處理和分析。We recommend using Azure Machine Learning text analytics modules for text processing and analysis.

Microsoft Web 語言模型 API 是 REST 架構的雲端服務,可提供最先進的工具來處理自然語言。The Microsoft Web Language Model API is a REST-based cloud service providing state-of-the-art tools for natural language processing. 使用此 API,應用程式便可透過 Bing 在 en-US 市場所收集的 Web 規模語料庫將語言模型定型,以利用巨量資料的強大威力。Using this API, your application can leverage the power of big data through language models trained on web-scale corpora collected by Bing in the en-US market.

這些平滑降速的 N-gram 語言模型會在下列語料庫上定型,最多可支援到第 5 階的 Markov 鏈結:These smoothed backoff N-gram language models, supporting up to fifth-order Markov chains, are trained on the following corpora:

  • 網頁本文Web page body text
  • 網頁標題文字Web page title text
  • 網頁錨定文字Web page anchor text
  • Web 搜尋查詢文字Web search query text

Web 語言模型 API 支援四個查閱作業:The Web Language Model API supports four lookup operations:

  1. 文字序列的聯結 (log10) 機率。Joint (log10) probability of a sequence of words.
  2. 指定了前面字組序列的單一文字條件式 (log10) 機率。Conditional (log10) probability of one word given a sequence of preceding words.
  3. 最可能遵循指定文字順序的文字清單 (完成)。List of words (completions) most likely to follow a given sequence of words.
  4. 不包含空格的字串斷字。Word breaking of strings that contain no spaces.

開始使用Getting Started

  1. 訂閱服務。Subscribe to the service.
  2. 下載 SDKDownload the SDK.
  3. 執行 SDK 程式碼範例。Run the SDK sample code.
  4. 請參閱 API 參考以取得端點的完整詳細資料,包括各種語言的程式碼片段。Refer to the API Reference for full details of the endpoints, including code snippets in a variety of languages.

基礎技術Underlying Technology

下列文件詳述這些語言模型的開發,使用這項服務的研究出版品中應該會加以引用:The following paper provides details on the development of these language models, and should be cited in research publications that use this service:

按一下此處來取得引用此著作的最新文件清單。Click here for a current list of papers citing this work.