Researchers offer insights on how diet ultimately reshapes language
Jan 2021, phys.org
Everett spent several years studying how environmental factors such as ambient aridity—extreme dryness—shift speech patterns by reducing vowel usage, which requires more effort to pronounce....Labiodental sounds such as "f" and "v"—sounds common today but rarely existed until soft diets became pervasive...In studying thousands of languages, the researchers established two linguistic camps—hunter-gatherers, whose diets have changed little and whose mouths get a lot more wear, and non-hunter-gatherers.via University of Miami: Caleb Everett et al, Speech adapts to differences in dentition within and across populations, Scientific Reports (2021). DOI: 10.1038/s41598-020-80190-8
Image credit: Mouthbreather, Habib M’henni
Shrinking massive neural networks used to model language
Dec 2020, phys.org
Chen and colleagues sought to pinpoint a smaller model concealed within BERT [a deep language model]. They experimented by iteratively pruning parameters from the full BERT network, then comparing the new subnetwork's performance to that of the original BERT model. They ran this comparison for a range of NLP [natural language processing] tasks, from answering questions to filling the blank word in a sentence.The researchers found successful subnetworks that were 40 to 90 percent slimmer than the initial BERT model, depending on the task.via Massachusetts Institute of Technology: Tianlong Chen et al. The Lottery Ticket Hypothesis for Pre-trained BERT Networks. arXiv:2007.12223 [cs.LG] arxiv.org/abs/2007.12223
AI can predict Twitter users likely to spread disinformation before they do it
Dec 2020, phys.org
Results from the study found that the Twitter users who shared stories from unreliable sources are more likely to tweet about either politics or religion and use impolite language. They often posted tweets with words such as 'liberal," 'government," 'media," and their tweets often related to politics in the Middle East and Islam, with their tweets often mentioning "Islam' or "Israel."In contrast, the study found that Twitter users who shared stories from reliable news sources often tweeted about their personal life, such as their emotions and interactions with friends. This group of users often posted tweets with words such as"mood." "wanna," "gonna," "I'll," "excited," and "birthday."via Uiversity of Sheffield: Identifying Twitter users who repost unreliable news sources with linguistic information, Yida Mu, Nikolaos Aletras, PeerJ, doi.org/10.7717/peerj-cs.325
After reading this, it's kind of ironic to think that social media is used to target people who have no social life. Then again, robots don't have much of a social life either... .
The linguistic device that creates resonance between people and ideas
Jan 2021, phys.org
In literature, writers often use the word "you" generically to make an idea seem more universal, even though it might not be....They found that highlighted passages [selected by electronic book readers] were 8.5 times more likely to contain generic "you" than passages that were not highlighted, leading them to identify generic-you as a linguistic device that enhances resonance."This study is a really nice example of how sensitive people are to even a subtle variation in perspective and language," Gelman said. "I'm sure people who are reading these novels were not thinking about the linguistic device the authors were using, and the authors themselves may not have been aware, but this study shows this linguistic device has a measurable effect, and that it's part of the fabric of language and thought that people are sensitive to."via University of Michigan: Ariana Orvell et al. "You" speaks to me: Effects of generic-you in creating resonance between people and ideas, Proceedings of the National Academy of Sciences (2020). DOI: 10.1073/pnas.2010939117
Linguists predict unknown words using language comparison
May 2021, phys.org
They found missing pieces in their dataset, but came up with a predictive algorithm to guess what they were. To test their predictions, they sought out native speakers and asked them about the missing words in the data. Turns out they got a 76% hit rate. (In this case, the dataset is 8 Western Kho-Bwa linguistic varieties spoken in India about which not much scholarship is documented.)
via the Max Planck Society: Timotheus A. Bodt et al, Reflex prediction, Diachronica (2021). DOI: 10.1075/dia.20009.bod
