For computers to recognize speech, they need to be provided with both text and audio. For popular languages like English, there is a wealth of both text and audio data. For lesser-known languages, text data might exist, but audio data is likely to be sparse. This means that of about 8,000 languages spoken across the world, most speech recognition technologies can only recognize 200 languages.

Researchers in Carnegie Mellon's Language Technologies Institute (LTI) hope to bring speech recognition to lesser-known languages by utilizing the similarities between commonly used languages and lesser-known ones. Most speech recognition technologies focus on phonemes, sounds that distinguish one word from a similar-sounding word, like the "d" in "dog" as opposed to the "l" in "log." However, the LTI researchers hope to utilize phones, which are sounds that are physically produced by language speakers, but don't necessarily differentiate between words.

The researchers are creating a language-approximation tool which analyzes how phones are shared between languages. This could reduce the need for audio data on lesser-known languages and allow models to recognize speech for up to 2000 languages. Currently, their model improves language-approximation tools by about 5 percent, but while they continue to improve the model, they hope that their work could serve as inspiration for other researchers.

In a paper published in "Science," Civil & Environmental Engineering professor David Rounce and other international collaborators estimate that in a worst-case scenario, 40 percent of the glacial mass will be gone within the century, accounting for 80 percent of the world's glaciers. In a best-case scenario, over 25 percent of glacial mass will be gone, accounting for 50 percent of glaciers by number. These projections come from a model using shared socioeconomic pathways, data on mass changes for individual glaciers, and calibrations that require the use of supercomputers. Rounce is also working on accounting for different types of glaciers such as tidewater or debris-covered glaciers, which can influence the rate of glacial melt. The researchers hope that nations will make more ambitious climate pledges and lower the 2.7 degree Celsius global temperature goal.

Krzysztof Matyjaszewski, J.C. Warner University Professor of Natural Sciences, will receive the National Academy of Sciences Award in Chemical Sciences in April for his advancements in polymer chemistry. He is recognized for discovering and developing atom transfer radical polymerization (ATRP), a method of forming polymers that allows scientists to easily control their molecular architecture. ATRP has been used in inkjet printing, cosmetics, packing material, adhesives, and could potentially be used for drug delivery, scaffolding for bone regeneration, degradable plastics, and other applications.

Theresa Mayer, former executive vice president for research and partnerships at Purdue University, will join Carnegie Mellon's executive management team in February as the Vice President of Research.

At Purdue, Mayer helped achieve record growth in external funding for basic and use-inspired research and diversified its portfolio of major sponsors. Mayer comes from a background in advanced manufacturing of nanoscale electronic, optical and biomedical devices, and has advised government councils and committees on science, technology, and policy issues.

If you're in the tech field, on the Internet, or even remotely social, you've probably heard about ChatGPT, which has taken the internet by storm since it was fully released to the public in November 2022.

It is safe to say that ever since, ChatGPT has spawned both panic and fascination. Developers, analysts, and writers marvel at its ability to accomplish tasks they perform in their jobs. Some are worried that it signals that AI is beginning to take away white-collar jobs. Others believe that AI can augment human work, ushering in a wave of new opportunities and increased productivity.

AI-enabled systems can also be used suboptimally. For example, many marketing teams use machine learning for data analytics, identifying users that are likely to stop viewing social media or website pages in order to prevent them from leaving. However, studies show that focusing efforts on users that are about to leave the page might not lead to higher user retention.

This new wave of AI and AI-enabled systems both excites and merits further consideration about the future of AI, education, work, and society.

While those who see the benefits of flu shots may find the proposal appealing, some scientists and FDA advisors found the proposal surprising.

For one, scientists don't feel that there is enough evidence suggesting civilians need annual doses. Some feel that they would like more information on who is most vulnerable to the virus and tailor the vaccine plan to them.

The FDA must also consider which vaccine to administer. The coronavirus has evolved quickly and is expected to continue creating new variants. The FDA has proposed that these vaccines be bivalent, meaning that they contain strains of the original virus and newer variants. However, studies have shown that bivalent vaccines may not be as effective as monovalent vaccines, because the body is more inclined to recognize the original virus rather than the new virus. Some experts feel that offering only monovalent vaccines tailored to the new viruses would be more effective. Some also suggest simplifying the vaccine to only target specific parts of the coronavirus.