Microsoft Technology Licensing, LLC

Kangzhiwei Zhejiang Technology Co., Ltd.

Furu Wei

Ming Zhou

Lei Cui

Shaohan Huang

Ji Li

Amit Srivastava

Xingxing Zhang

Li Dong

Yang Liu

Nanshan Zeng

Shujie Liu

Yu Wu

Xuedong David Huang

Chengyi Wang

Kenichi Kumatani

Ziqiang Cao

Yao Qian

Wei-Ying Ma

Zhen Liu

Jian Luan

Tomasz Lukasz Religa

Kaushik Ramaiah Narayanan

Max Wang

Kati Amanda London

Huitian Jiao

Si-Qing Chen

Nan Yang

Qinying Liao

Ravi Mandliya

Haoyong Zhang

Tao Ge

Utsab Bose

Xiangqian Chen

Yu-Ting Kuo

There is provided a solution for semantic representation of text in a document. In this solution, textual information comprising a sequence of text elements () and layout information () of the text element are determined from a document. The layout information () indicates a spatial arrangement of the plurality of text elements () presented within the document. Based at least in part on the plurality of text elements () and the layout information (), respective semantic feature representations () of the plurality of text elements () are generated. By jointly using both the textual information and the layout information (), rich semantics of the text elements () in the document can be effectively captured in the feature representations.

Semantic representation of text in document

A system obtains a first training data set comprising labeled speech data or both labeled and unlabeled data corresponding to a high-resource data set as well as latent speech representations based on the first training data set. The system trains a machine learning model on the first training data set to learn phonetically aware speech representations corresponding to the first training data set. The system applies the latent speech representations to a transformer context network to generate contextual representations. The system aligns each of the contextual representations with a phoneme label to generate phonetically-aware contextual representations. The system causes a refinement engine to further refine the machine learning model.

Unified speech representation learning

A data processing system implements obtaining first textual content in a first language from a first client device; determining that the first language is supported by a first machine learning model; obtaining a guard list of prohibited terms associated with the first language; determining that the textual content does not include one or more prohibited terms associated based on the guard list; providing the first textual content as an input to the first machine learning model responsive to the textual content not including the one or more prohibited terms; analyzing the first textual content with the first machine learning model to obtain a first content recommendation; obtaining a first content recommendation policy that identifies content associated with the first language that may not be provided as a content recommendation; determining that the first content recommendation is not prohibited; and providing the first content recommendation to the first client device.

Multilingual content recommendation pipeline

A system and method for converting a document is described. The system accesses a document comprising one or more section breaks. The system detects sections of the text document demarked by the one or more section breaks and generates a section title metadata and a section summary metadata for each section of the plurality of sections. The system inserts the section title metadata and the section summary metadata at the corresponding section breaks in the text document. The system modifies the text document into slides. Each slide being formed for each section based on the corresponding section title metadata and the section summary metadata. The system generates a presentation document based on the slides.

Document conversion engine

The present disclosure provides a technical solution of live comments generating, which may acquire candidate texts highly similar with segments of video as live comments of corresponding segments by matching the candidate texts with the segments, and further generate new live comments based on video segments and existed live comments to enrich the live comments information of related video.

Live comments generating

Systems and methods are provided for training a machine learning model to learn speech representations. Labeled speech data or both labeled and unlabeled data sets is applied to a feature extractor of a machine learning model to generate latent speech representations. The latent speech representations are applied to a quantizer to generate quantized latent speech representations and to a transformer context network to generate contextual representations. Each contextual representation included in the contextual representations is aligned with a phoneme label to generate phonetically-aware contextual representations. Quantized latent representations are aligned with phoneme labels to generate phonetically aware latent speech representations. Systems and methods also include randomly replacing a sub-set of the contextual representations with quantized latent speech representations during their alignments to phoneme labels and aligning the phonetically aware latent speech representations to the contextual representations using supervised learning.

A method and system for training a text-to-content recommendation ML model includes training a first ML model using a first training data set, utilizing the trained first ML model to infer information about the data contained in the first training data set, collecting the inferred information to generate a second training data set, and utilizing the first training data set and the second training data set to train a second ML model. The second ML model may be a text-to-content recommendation ML model.

Cross data set knowledge distillation for training machine learning models

Implementations of the subject matter described herein provide a solution that enables a machine to automatically generate a melody. In this solution, user emotion and/or environment information is used to select a first melody feature parameter from a plurality of melody feature parameters, wherein each of the plurality of melody feature parameters corresponds to a music style of one of a plurality of reference melodies. The first melody feature parameter is further used to generate a first melody that conforms to the music style and is different from the reference melody. Thus, a melody that matches user emotions and/or environmental information may be automatically created.

Autonomous generation of melody

Methods and systems for linking comments to portions of content items. An example computing device receives information associated with a content item produced by a source system, the content item being accessible to other the computing devices via a network and receives a comment associated with the content item, the comment produced by one of the other computing devices. In response to receiving the information and the comment, the computing device predicts a subsection of the content item to link to the received comment based at least on details associated with the content item and the comment, then makes information associated with the predicted subsection of the content item available to other computing devices requesting access to the content item.

Systems and methods for providing a comment-centered news reader

A method and system for providing an application-specific embedding for an entire text-to-content suggestions service is disclosed. The method includes accessing a dataset containing unlabeled training data collected from an application, the unlabeled training data being collected under user privacy constraints, applying an unsupervised ML model to the dataset to generate a pretrained embedding; and utilizing the pretrained embedding to train the text-to-content suggestion ML model utilized by the application.

Method and system of utilizing unsupervised learning to improve text to content suggestions

Method and system for training a text-to-content suggestion ML model include accessing a dataset containing unlabeled training data collected from an application, the unlabeled training data being collected under user privacy constraints, applying an ML model to the dataset to generate a pretrained embedding, and applying a supervised ML model to a labeled dataset to train the text-to-content suggestion ML model utilized by the application by utilizing the pretrained embedding generated by the supervised ML model.

An example environment can automatically generate content for a conversation in response to a presented content item. An example computing device receives information associated with a content item produced by a source system, the content item being accessible to other the computing devices via a network. In response to receiving the information, the computing device identifies a start comment. In response to identifying the start comment, the computing device identifies a reply to the start comment using at least a ranking model.

Automatic conversation creator for news

In accordance with implementations of the subject matter described herein, there is provided a solution for supporting a machine to automatically generate a song. In this solution, an input from a user is used to determine a creation intention of the user with respect to a song to be generated. Lyrics of the song are generated based on the creation intention. Then, a template for the song is generated based at least in part on the lyrics. The template indicates a melody matching with the lyrics. In this way, it is feasible to automatically create the melody and lyrics which not only conform to the creation intention of the user but also match with each other.

Automatic song generation

Methods and systems for providing a comments-centered news reader. Configurations allow live comments to be presented along with the news or similar website content. While a user scrolls up and down in a browser presenting a news article on the user's computer device (e.g., mobile device), linked comments are shown in a selected region. The displayed comments automatically change to adapt to what parts (paragraphs, sentences) of the news article that user is currently reading. At the same time, users can publish their own comments without having to proceed to a separate section of the browser, thus saving the viewer actions and improving the user's experience. The user's system or a remote server records the comments along with the article or the place users are in the article when the comment was entered.

Comment-centered news reader

A multi-layer water sealing device, includes at least two layers of water seal sequentially arranged from top to bottom. An upper-layer water seal has a floating bell-shaped cover structure, and the floating bell-shaped cover structure includes a water sealing box and a floating bell-shaped cover arranged in the water sealing box. A lower-layer water seal is arranged in a drain pipe below the floating bell-shaped cover, and one side of the water sealing box is extended to form or connected to a communication cavity offset from the drain pipe. The floating bell-shaped cover correspondingly is extended to form or connected to a floating component floatable in the communication cavity and offset from the drain pipe.

Growing community of inventors

Beijing, China

Furu Wei