Word of Mouth: Introducing Voice Search for Indonesian, Malaysian and Latin American Spanish

Posted by Linne Ha, International Program Manager

Read more about the launch of Voice Search in Latin American Spanish on the Google América Latina blog.

Today we are excited to announce the launch of Voice Search in Indonesian, Malaysian, and Latin American Spanish, making Voice Search available in over two dozen languages and accents since our first launch in November 2008. This accomplishment could not have been possible without the help of local users in the region - really, we couldn’t have done it without them. Let me explain:

In 2010 we launched Voice Search in Dutch, the first language where we used the “word of mouth” project, a crowd-sourcing effort to collect the most accurate voice data possible.The traditional method of acquiring voice samples is to license the data from companies who specialize in the distribution of speech and text databases. However, from day one we knew that to build the most accurate Voice Search acoustic models possible, the best data would come from the people who would use Voice Search once it launched - our users.

Since then, in each country, we found small groups of people who were avid fans of Google products and were part of a large social network, either in local communities or on online. We gave them phones and asked them to get voice samples from their friends and family. Everyone was required to sign a consent form and all voice samples were anonymized. When possible, they also helped to test early versions of Voice Search as the product got closer to launch.

Building a speech recognizer is not just limited to localizing the user interface. We require thousands of hours of raw data to capture regional accents and idiomatic speech in all sorts of recording environments to mimic daily life use cases. For instance, when developing Voice Search for Latin American Spanish, we paid particular attention to Mexican and Argentinean Spanish. These two accents are more different from one another than any other pair of widely-used accents in all of South and Central America. Samples collected in these countries were very important bookends for building a version of Voice Search that would work across the whole of Latin America. We also chose key countries such as Peru, Chile, Costa Rica, Panama and Colombia to bridge the divergent accent varieties.

As an International Program Manager at Google, I have been fortunate enough to travel around the world and meet many of our local Google users. They often have great suggestions for the products that they love, and word of mouth was created with the vision that our users could participate in developing the product. These Voice Search launches would not have been possible without the help of our users, and we’re excited to be able to work together on the product development with the people who will ultimately use our products.

Reading tea leaves in the tourism industry: A Case Study in the Gulf Oil Spill

Posted by Hyunyoung Choi and Paul Liu, Senior Economists

A few years ago, our in-house economists, Hal Varian and Hyunyoung Choi, demonstrated how to “predict the present” with monthly visitor arrivals to Hong Kong. We took this idea further to see if search queries could predict the future. If users start to research their travel plans some weeks or months in advance, then intuitively shouldn’t we be able to extend "predicting the present" into "predicting the future?" We decided to test it out by focusing on a region whose tourism was recently severely impacted: Florida’s gulf coast.

With the travel industry still in the midst of recovering from a deep recession, the Gulf Oil spill had the potential to do significant economic damage. Our case study on the Gulf Oil spill helped find useful insight into people’s future travel plans to Florida; in fact, we found that travel search queries actually were good predictors for trips to Florida, and destinations within Florida, about 4 weeks later.

The results we saw surprised us. Google Insights for Search suggested that at least with respect to hotel bookings (using data from Smith Travel Research, Inc.), the aggregate effect of the oil spill was modest on Florida travel, since travelers tended to shift their destinations from the affected regions on the west coast to the east coast or central regions of Florida. In particular, hotel bookings for affected areas along the Gulf coast were 4.25% less than predicted, and unaffected areas along the Atlantic coast were 4.89% greater than predicted.

You can read the full case study here or try your own hand at predicting the future!

Games, auctions and beyond

Posted by Yossi Matias, Senior Director, Head of Israel R&D Center

In an effort to advance the understanding of market algorithms and Internet economics, Google has launched an academic research initiative focused on the underlying aspects of online auctions, pricing, game-theoretic strategies, and information exchange. Twenty professors from three leading Israeli academic institutions - the Hebrew University, Tel Aviv University and the Technion - will receive a Google grant to conduct research for three years.

In the past two decades, we have seen the Internet grow from a scientific network to an economic force that positively affects the global economy. E-commerce, online advertising, social networks and other new online business models present fascinating research questions and topics of study that can have a profound impact on society.

Consider online advertising, which is based on principles from algorithmic game theory and online auctions. The Internet has enabled advertising that is more segmented and measurable, making it more efficient than traditional advertising channels, such as newspaper classifieds, radio spots, and television commercials. These measurements have led to better pricing models, which are based on online real-time auctions. The original Internet auctions were designed by the industry, based on basic economic principles which have been known and appreciated for forty years.

As the Internet grows, online advertising is becoming more sophisticated, with developments such as ad-exchanges, advertising agencies which specialize in online markets, and new analytic tools. Optimizing value for advertisers and publishers in this new environment may benefit from a better understanding of the strategies and dynamics behind online auctions, the main driving tool of Internet advertising.

These grants will foster collaboration and interdisciplinary research by bringing together world renowned computer scientists, engineers, economists and game theorists to analyze complex online auctions and markets. Together, they will help bring this area of study into mainstream academic scientific research, ultimately advancing the field to the benefit of the industry at large.

The professors who received research grants include:

• Hebrew University: Danny Dolev, Jeffrey S. Rosenschein, Noam Nisan (Computer Science and Engineering); Liad Blumrosen, Alex Gershkov, Eyal Winter (Economics); Michal Feldman and Ilan Kremer (Business). The last six are also members of the Center for the Study of Rationality.
• Tel Aviv University: Yossi Azar, Amos Fiat, Haim Kaplan, and Yishay Mansour (Computer Science); Zvika Neeman (Economics); Ehud Lehrer and Eilon Solan (Mathematics); and Gal Oestreicher (Business).
• Technion: Seffi Naor (Computer Science); Ron Lavi (Industrial Engineering); Shie Mannor and Ariel Orda (Electrical Engineering).

In addition to providing the funds, Google will offer support by inviting the researchers to seminars, workshops, faculty summits and brainstorming events. The results of this research will be published for the benefit of the Internet industry as a whole, and will contribute to the evolving discipline of market algorithms.

Large Scale Image Annotation: Learning to Rank with Joint Word-Image Embeddings

Posted by Jason Weston and Samy Bengio, Research Team

In our paper, we introduce a generic framework to find a joint representation of images and their labels, which can then be used for various tasks, including image ranking and image annotation.

We focus on the task of automatic assignment of annotations (text labels) to images given only the pixel representation of the image (i.e., with no known metadata). This is achieved by a learning algorithm, that is, where the computer learns to predict annotations for new images given annotated training images. Such training datasets are becoming larger and larger, with tens of millions of images and tens of thousands of possible annotations. In this paper, we propose a strongly performing method that scales to such datasets by simultaneously learning to optimize precision at the top of the ranked list of annotations for a given image and learning a low-dimensional joint embedding vector space for both images and annotations. Our system learns an interpretable model, where annotations with alternate wordings ("president obama" or "barack"), different languages ("tour eiffel" or "eiffel tower"), or similar concepts (such as "toad" or "frog") are close in the embedding space. Hence, even when our model does not predict the exact annotation given by a human labeler, it often predicts similar annotations.

Our system is trained on ~10 million images with ~100,000 possible annotation types and it annotates a single new image in ~0.17 seconds (not including feature processing) and consumes only 82MB of memory. Our method both outperforms all the methods we tested against and in comparison to them is faster and consumes less memory, making it possible to house such a system on a laptop or mobile device.

Building resources to syntactically parse the web

Posted by Slav Petrov and Ryan McDonald, Research Team

One major hurdle in organizing the world’s information is building computer systems that can understand natural, or human, language. Such understanding would advance if systems could automatically determine syntactic and semantic structures.

This analysis is an extremely complex inferential process. Consider for example the sentence, "A hearing is scheduled on the issue today." A syntactic parser needs to determine that "is scheduled" is a verb phrase, that the "hearing" is its subject, that the prepositional phrase "on the issue" is modifying the "hearing", and that today is an adverb modifying the verb phrase. Of course, humans do this all the time without realizing it. For computers, this is non-trivial as it requires a fair amount of background knowledge, typically encoded in a rich statistical model. Consider, "I saw a man with a jacket" versus "I saw a man with a telescope". In the former, we know that a "jacket" is something that people wear and is not a mechanism for viewing people. So syntactically, the "jacket" must be a property associated with the "man" and not the verb "saw", i.e., I did not see the man by using a jacket to view him. Whereas in the latter, we know that a telescope is something with which we can view people, so it can also be a property of the verb. Of course, it is ambiguous, maybe the man is carrying the telescope.


Linguistically inclined readers will of course notice that this parse tree has been simplified by omitting empty clauses and traces.

Computer programs with the ability to analyze the syntactic structure of language are fundamental to improving the quality of many tools millions of people use every day, including machine translation, question answering, information extraction, and sentiment analysis. Google itself is already using syntactic parsers in many of its projects. For example, this paper, describes a system where a syntactic dependency parser is used to make translations more grammatical between languages with different word orderings. This paper uses the output of a syntactic parser to help determine the scope of negation within sentences, which is then used downstream to improve a sentiment analysis system.

To further this work, Google is pleased to announce a gift to the Linguistic Data Consortium (LDC) to create new annotated resources that can facilitate research progress in the area of syntactic parsing. The primary purpose of the gift is to generate data sets that language technology researchers can use to evaluate the robustness of new parsing methods in several web domains, such as blogs and discussion forums. The goal is to move parsing beyond its current focus on carefully edited text such as print news (for which annotated resources already exist) to domains with larger stylistic and topical variability (where spelling errors and grammatical mistakes are more common).

The Linguistic Data Consortium is a non-profit organization that produces and distributes linguistic data to researchers, technology developers, universities and university libraries. The LDC is hosted by the University of Pennsylvania and directed by Mark Liberman, Christopher H. Browne Distinguished Professor of Linguistics.

The LDC is the leader in building linguistic data resources and will annotate several thousand sentences with syntactic parse trees like the one shown in the figure. The annotation will be done manually by specially trained linguists who will also have access to machine analysis and can correct errors the systems make. Once the annotation is completed, the corpus will be released to the research community through the LDC catalog. We look forward to seeing what they produce and what the natural language processing research community can do with the rich annotation resource.

Slicing and dicing data for interactive visualization

Posted by Benjamin Yolken, Google Public Data Product Manager

A year ago, we introduced the Google Public Data Explorer, a tool that allows users to interactively explore public-interest datasets from a variety of influential sources like the World Bank, IMF, Eurostat, and the US Census Bureau. Today, users can visualize over 300 metrics across 31 datasets, including everything from labor productivity (OECD) to Internet speed (Ookla) to gender balance in parliaments (UNECE) to government debt levels (IMF) to population density by municipality (Statistics Catalonia), with more data being added every week.

Last week, as part of the launch of our dataset upload interface, we released one of the key pieces of technology behind the product: the Dataset Publishing Language (DSPL). We created this format to address a key problem in the Public Data Explorer and other, similar tools, namely, that existing data formats don’t provide enough information to support easy yet powerful data exploration by non-technical users.

DSPL addresses this by adding an additional layer of metadata on top of the raw, tabular data in a dataset. This metadata, expressed in XML, describes the concepts in the dataset, for instance “country”, “gender”, “population”, and “unemployment”, giving descriptions, URLs, formatting properties, etc. for each. These concepts are then referenced in slices, which partition the former into dimensions (i.e., categories) and metrics (i.e., quantitative values) and link them with the underlying data tables (provided in CSV format). This structure, along with some additional metadata, is what allows us to provide rich, interactive dataset visualizations in the Public Data Explorer.

With the release of DSPL, we hope to accelerate the process of making the world’s datasets searchable, visualizable, and understandable, without requiring a PhD in statistics. We encourage you to read more about the format and try it yourself, both in the Public Data Explorer and in your own software. Stay tuned for more DSPL extensions and applications in the future!