NEW YORK - Unsubstantiated rumours and outright falsehoods spread widely in immigrant communities before the US presidential election in 2020.

That is happening again in the run-up to 2022's midterm elections, researchers say, but with an insidious twist: The social media accounts pushing misinformation are now targeting audiences in more languages on more topics and across more digital platforms, with scant resistance from social media companies.

In recent weeks, posts exaggerating the fallout from inflation have been aimed at Americans from Latin American countries that have been crippled by poor economic management. Conspiracy theories that spread in August about the IRS's plans for a "shadow army" led mentions of "Ejército IRS" to surge alongside "IRS army", its equivalent in English, according to the research group Zignal.

Misinformation swirling in Chinese on Twitter, YouTube and WeChat about mail-in ballots, school curricula and hate crimes "has dangerous implications" this year for Asian American voters, who are growing as a political force, according to the advocacy group Asian Americans Advancing Justice.

"There's definitely a hyper-targeting of messaging," said Mr Nick Nguyen, a co-founder of Viet Fact Check, a group that offers explanations about misinformation circulating among Vietnamese Americans. "This is where a lack of English-language fluency can make populations vulnerable."

Viet Fact Check is among a growing number of groups trying to contextualise and debunk false online narratives in languages other than English.

Factchequeado, a six-month-old Spanish-language service, is examining inaccurate translations, manipulated images and misleadingly edited videos about the search of Mar-a-Lago and House Speaker Nancy Pelosi's visit to Taiwan.

Desifacts, which focuses on South Asian American communities, began publishing explainers and clarifications about topics such as immigration and student debt relief in Hindi, Bengali and Tamil in February.

But the multilingual fact-checkers say they cannot keep pace with the deluge of falsehoods online. They have called on the big social media platforms, such as Facebook and YouTube, to do more for efforts in other languages as they would for misinformation in English.

"With mis- and disinformation in Spanish, we feel like we are fighting a giant," said Tamoa Calzadilla, Factchequeado's managing editor and the former head of Univision's fact-checking operation. "It's frustrating because we are trying to do something, and we need support from the platforms - we are doing our work, but Big Tech can do more."

The social media companies said they moderated content or provided fact-checks in many languages: more than 70 languages for TikTok, and more than 60 for Meta, which owns Facebook. YouTube said it had more than 20,000 people reviewing and removing misinformation, including in languages such as Mandarin and Spanish; TikTok said it had thousands.

The companies declined to say how many employees were doing work in languages other than English.

TikTok has translated a midterms information hub on its app into more than 45 languages. Twitter has a similar elections centre available in English and Spanish, along with prompts that debunk and "pre-bunk" misinformation in different languages. Meta said that it had invested in initiatives such as Spanish fact-checking services on WhatsApp in preparation for the elections and that it would show voting-related notifications in both English and a second language based on users' activity.

The companies also cited broader improvements.

Meta said its Spanish misinformation prediction models in the United States were now working on a par with its English-language models and had significantly increased the amount of Spanish content sent to fact-checkers for review. Twitter said its newly redesigned contextual labels, which are translated based on users' language settings, had helped shrink engagement with misinformation. YouTube said information panels now appeared in different languages for certain search results and videos. It also highlights content from vetted non-English news sources based on users' language settings and search queries, the company said.

But researchers worry about the effect of non-English misinformation on the coming vote, saying lies and rumours in other languages continue to seep through.

A report released Monday from the watchdog group Media Matters found 40 Spanish-language videos on YouTube that advanced misinformation about US elections, including the false notion that fraudulent ballots were coming into the United States from China and Mexico.

Some disinformation experts, along with some elected officials, have pressed the social platforms for more action and transparency.

This year, the Congressional Hispanic Caucus pushed Meta, TikTok, YouTube and Twitter for meetings with their top executives to discuss the spread of misinformation in Spanish. YouTube made its CEO Susan Wojcicki available; TikTok and Twitter sent other executives. The caucus and Meta were unable to schedule a meeting, and Meta said it planned to instead submit a written update.

The International Fact-Checking Network at Poynter sent an open letter to YouTube in January, describing the ease with which misinformation on the platform was flowing across borders. Researchers have said the same narrative often emerges on different sites in different countries, and then cross-pollinates in a feedback loop that makes it seem more believable.

As one fact-checker argued, an immigrant is more likely to trust a conspiracy theory voiced by both the person's mother in El Salvador and a friend in San Francisco.

Misinformation can also do what researchers call platform-jump - originate in English on fringe services like Truth Social or Gab and then emerge later on more mainstream sites, presented in a different language or sometimes with a misleading translation attached.

Some experts, sceptical that all multilingual misinformation can be removed, push instead for other ways to limit amplification. Last year, Twitter tested a feature that allowed some users in the United States, South Korea and Australia to flag tweets as misleading. NYTIMES