Google Translate is adding 110 new languages
Google’s PaLM 2 AI language model enabled Translate to learn these
Google Translate is adding 110 new languages, including Cantonese, NKo, and Tamazight, using the PaLM 2 language model.
This expansion, the largest ever for Google Translate, will support more than 614 million speakers, representing about 8% of the world’s population. Among the new languages, a significant portion comes from Africa.
Notable languages added include:
Afar: A tonal language spoken in Djibouti, Eritrea, and Ethiopia.
Cantonese: A highly requested language that overlaps with Mandarin in writing.
Manx: A revived Celtic language from the Isle of Man.
NKo: A standardized form of the West African Manding languages.
Punjabi (Shahmukhi): The Perso-Arabic script variety of Punjabi, primarily spoken in Pakistan.
Tamazight (Amazigh): A Berber language spoken across North Africa, supported in both Latin and Tifinagh scripts.
Tok Pisin: An English-based creole spoken in Papua New Guinea.
Google selected language varieties based on common usage, focusing on widely used dialects, and aims to continue expanding support for more languages.
Earlier this month, Meta announced an AI system that aims to support endangered languages by scaling up machine translation for languages with limited digital resources. The initiative involves translating over 200 languages.
Things I’m reading today
Nokia to acquire Infinera for $2.3 billion to expand its optical networking business (link)
Nokia plans to acquire Infinera Corp. for $2.3 billion to enhance its optical networking business. This acquisition, aimed at achieving double-digit operating margins, follows Nokia's sale of its undersea internet cable business. Nokia will pay $6.65 per share, a 28% premium, with 70% in cash and up to 30% in American depositary shares. The deal includes repurchasing Infinera's $760 million in convertible notes.
Nokia's CEO, Pekka Lundmark, highlighted the strategic decision to invest in optical networks in 2021, which has led to improved sales and profitability. The acquisition could expand Nokia's global scale in optical networks by 75%, accelerate its product roadmap, and enhance its technological capabilities. The deal is expected to boost Nokia's earnings per share in the first year and deliver over 10% EPS accretion by 2027. To counter dilution, Nokia's board will increase its share buyback program (Via Mike Wheatley/ SiliconANGLE).
Meta’s LLM Compiler is the latest AI innovation that aims to transform the way we code (link)
Meta has introduced the Meta Large Language Model (LLM) Compiler, an open-source suite designed to optimize code and innovate compiler design. Trained on 546 billion tokens of LLVM-IR and assembly code, the LLM Compiler understands compiler intermediate representations, assembly language, and optimization techniques, enabling it to perform tasks usually handled by human experts or specialized tools.
This AI-powered compiler achieves significant results in code size optimization, reaching 77% of the potential of autotuning searches, thereby reducing compilation times and improving code efficiency. It also demonstrated a 45% success rate in round-trip disassembly, valuable for reverse engineering and legacy code maintenance.
Chris Cummins, a core contributor, highlighted the model's impact, noting the release of pre-trained models in two sizes (7 billion and 13 billion parameters) and their effectiveness in fine-tuned versions. The technology offers benefits like faster compile times and more efficient code, while its permissive commercial license encourages both academic and industrial innovation (Via Michael Nuñez/VentureBeat).
