"Images are universal, but image search is not," said Oren Etzioni, a professor of computer science and engineering at the University of Washington. "A person who types his or her search in English won't find images tagged in Chinese, and a Dutch person won't find images tagged in English. We've created a collaborative tool that solves this problem."
A new multilingual search tool developed at the UW's Turing Center makes the universal appeal of pictures available to all. PanImages, presented today at the Machine Translation Summit in Copenhagen, Denmark, allows people to search for images on the Web using hundreds of languages.
Search engines such as Google look for images by detecting the search term in captions and other nearby text. But since the process looks for a string of letters, the results are limited to the seeker's mother tongue.
The new tool is named PanImages, from the Greek prefix, "pan," meaning whole or all-inclusive. It automatically translates the search term into about 300 other languages, suggests a few that might work and then displays images from Google and the online photo database Flickr.
PanImages promises to help people who speak languages that have a small Web presence. Imagine you are a Zulu speaker looking for a picture of a refrigerator, Etzioni said. You type the Zulu word for refrigerator ("ifriji") into an image search and get two results. The same search using PanImages generates 472,000 hits. In a test of so-called minor languages, PanImages was able to find 57 times more results, on average, than a Google image search.
"We want to serve the vast number of people who don't speak one of the major languages," Etzioni said. "As the Internet becomes more widely available outside of the major industrialized nations, it becomes increasingly important to serve people who don't speak English, French or Chinese."
Even people who speak these more common languages can benefit by switching electronic tongues. Words that have more than one meaning inevitably produce unwanted results. For instance, typing the word "spring" in an English-language image search generates diverse images: grassy meadows, metal coils and pictures from the town of Silver Spring, Md. If you want images of a metal spring, you might use PanImages and search for the more precise French word "ressort." If you want a picture of a rectangular bar and don't want businesses where patrons drink alcohol, you might try the Russian word "áðóñîê." Experiments showed that, for common languages, PanImages nearly doubles the number of correct images on the first 15 pages of results.
PanImages' powerful brains were created by scanning more than 350 machine-readable online dictionaries. Some of these were "wiktionaries," online multilingual dictionaries written by volunteers. The PanImages software scans these dictionaries and uses an algorithm to check the accuracy of the results. It then assembles the results in a matrix that allows translation in combinations that may never have been attempted -- for instance, from Gujarati to Lithuanian.
"It's an unprecedented lexical resource. The most distinguishing element is its ability to scale to such a broad set of languages," Etzioni said. "Our goal is to ultimately cover all the languages people are interested in."
Free online translation services used by Yahoo! and Google incorporate just one or two dozen common languages. In the United States, research on machine translation tends to focus on languages with military importance, such as Arabic and Chinese, Etzioni said. PanImages had 50 languages earlier this year and by June it incorporated 100 languages. It now includes some 300 languages, 2.5 million words and millions of individual translations.
PanImages also lets people instantly add new words or translations.
Future work on PanImages will scour more online dictionaries to expand the number of words and languages it can handle. Researchers also hope to translate the words used in tagging sites, such as del.icio.us, where visitors use single-word labels to describe the page's content.
"Our goal is to promote pan-lingual translation," said Etzioni. "With this first step, we've created a service we hope will be a handy tool."
Hannah Hickey | EurekAlert!
The plastic brain: Better connectivity of brain regions with training
02.07.2018 | Leibniz-Institut für Wissensmedien
Arguments, Emotions, and News distribution in social media - Leibniz-WissenschaftsCampus Tübingen
04.05.2018 | Leibniz-Institut für Wissensmedien
There are currently great hopes for solid-state batteries. They contain no liquid parts that could leak or catch fire. For this reason, they do not require cooling and are considered to be much safer, more reliable, and longer lasting than traditional lithium-ion batteries. Jülich scientists have now introduced a new concept that allows currents up to ten times greater during charging and discharging than previously described in the literature. The improvement was achieved by a “clever” choice of materials with a focus on consistently good compatibility. All components were made from phosphate compounds, which are well matched both chemically and mechanically.
The low current is considered one of the biggest hurdles in the development of solid-state batteries. It is the reason why the batteries take a relatively long...
New design tool automatically creates nanostructure 3D-print templates for user-given colors
Scientists present work at prestigious SIGGRAPH conference
Most of the objects we see are colored by pigments, but using pigments has disadvantages: such colors can fade, industrial pigments are often toxic, and...
Scientists at the University of California, Los Angeles present new research on a curious cosmic phenomenon known as "whistlers" -- very low frequency packets...
Scientists develop first tool to use machine learning methods to compute flow around interactively designable 3D objects. Tool will be presented at this year’s prestigious SIGGRAPH conference.
When engineers or designers want to test the aerodynamic properties of the newly designed shape of a car, airplane, or other object, they would normally model...
Researchers from TU Graz and their industry partners have unveiled a world first: the prototype of a robot-controlled, high-speed combined charging system (CCS) for electric vehicles that enables series charging of cars in various parking positions.
Global demand for electric vehicles is forecast to rise sharply: by 2025, the number of new vehicle registrations is expected to reach 25 million per year....
17.08.2018 | Event News
08.08.2018 | Event News
27.07.2018 | Event News
21.08.2018 | Ecology, The Environment and Conservation
21.08.2018 | Life Sciences
21.08.2018 | Power and Electrical Engineering