Google open software source code language: a program comparable to human linguist-Google, the source, language-IT information

May 13 morning news, Google today opened a natural language understanding software source code for SyntaxNet, which as part of the company TensorFlow open source machine learning library. Meaning of this software can be used to automatically parse the statement, published by the code include the training of the new model, as well as English training model of text analysis.

Google said it named Parsey McParseface parser will automatically determine whether a Word is a noun, verb or adjective, it is currently the world's most accurate program of its kind, comparable to and even human linguists.

This technique has an extremely important meaning in natural language research. For Google itself, but also of great significance.

"Our internal assessment approach is very different. We don't care much about benchmarks, more concerned about the impact on the performance of downstream systems. Our goal is to improve the user experience. "Google product managers of the Institute daifu·aoer (Dave Orr) said.

As with TensorFlow, SyntaxNet mainly use C++ implementation. It is now open source, so outside programmers can also be improved, so as to help the company find new talent and improve products. Overall, statement analysis and product reviews, and includes comments as well as restaurants and shopping reviews, the technology and Internet search and Google Now On Tap are also related to the function.

"This is very important, because the language is sometimes subtle, may not be able to directly understand what people mean, very close to some content and context. "Google's head of Research Institute taniya·baidelakesi-Weiss (Tania Bedrax-Weiss) said.

Orr said that compared with the traditional machine learning algorithms, and more at depth of learning technologies in language comprehension. This method usually requires a large amount of data on the artificial neural network is trained, and extrapolating from the new data. Google will also deep learning technology for image recognition and speech recognition. In fact, the neural network is the key to SyntaxNet, the development of the project, code-named "neurosis" (neurosis).

谷歌开放语言软件源代码：程序媲美人类语言学家 - 谷歌,源代码,语言 - IT资讯

5月13日上午消息，谷歌今天开放了自然语言理解软件SyntaxNet的源代码，将其作为该公司TensorFlow开源机器学习库的一部分。这款软件可以用于自动分析语句含义，而此次公布的包括训练新模型的代码，以及英语文本分析的预训练模型。

谷歌表示，这个名为Parsey McParseface的句法分析程序可以自动判断某个单词是名词、动词还是形容词，它是目前全球同类程序中准确度最高的一款，甚至可以与人类语言学家媲美。

这种技术在自然语言研究领域拥有极其重要的意义。但对谷歌本身而言同样意义重大。

“我们内部评估技术的方法非常不同。我们不太关心基准，更加关心对下游系统性能的影响。我们的目标是改善用户体验。”谷歌研究院产品经理戴夫·奥尔(Dave Orr)说。

与TensorFlow一样，SyntaxNet主要使用C++执行。它现在实现了开源，使得外部程序员也可以对其加以改进，从而帮助该公司寻找新的人才并改进产品。整体而言，语句分析与产品评论有关，包括应用评论以及餐馆和购物点评，这项技术与互联网搜索和Google Now On Tap功能也有关系。

“这非常重要，因为语言有的时候很微妙，未必能直接理解人们的意思，有些内容与上下文关系很紧密。”谷歌研究院团队主管塔尼亚·拜德拉克斯-维斯(Tania Bedrax-Weiss)说。

奥尔表示，与传统的机器学习算法相比，深度学习技术在语言理解方面更加擅长。这种方法通常需要通过大量数据对人工神经网络进行训练，然后让其对新数据进行推断。谷歌还将深度学习技术用于图片识别和语音识别。事实上，神经网络是SyntaxNet的关键所在，该项目的开发代号为“神经官能症”(neurosis)。