More Gemini Features Coming to Google TV
Google TV integrates Gemini AI for image/video generation, enhances Google Photos, and adds a dedicated YouTube Shorts feed to the homepage.
Read moreKross-language learning is an AI technique that involves training a natural language processing model in one language and then training it in another language. It has been proven that retrained models can outperform models trained from scratch using the second language, which is why researchers at Amazon's Alexa division have invested a significant amount of time in studying this technique.
In a paper to be presented at this year's Conference on Empirical Methods in Natural Language Processing, two scientists from Alexa AI's Natural Language Understanding team, Quynh Do and Judith Gaspers, along with their colleagues, have developed a data selection technique that reduces the required training data by half. They claim that this technique has successfully improved the overall performance of the model in the target language.
Do and Gaspers wrote in a blog post, "Sometimes the data in the source language is so abundant that it is impractical to use all of it to train the transfer model. Additionally, the linguistic differences between the source and target languages mean that the training data in the source language needs to be refined to better match the statistical patterns of the target language."During the experiment, Do, Gaspers, and their team used two methods to split the source language dataset in half: data selection technique and random sampling. They pretrained individual models on both the halved datasets and the entire dataset, and then fine-tuned the models on a smaller dataset.
Do and Gaspers noted that all models were trained simultaneously to recognize intents (actions requested) and fill slots (variables acted upon by the intents), and they used multilingual embeddings as input (mapping a word or word sequence from different languages to a single point in a multidimensional space) to improve the accuracy of the model. The team combined the multilingual embeddings of each input word with character-level embeddings (encoding information about the prefixes, suffixes, and roots of the word), and they utilized a language model trained on a large text corpus to select the data to be fed into the source language transfer model.In the system designed by the researchers, a bilingual dictionary converts each pronunciation in the source dataset into a string of characters in the target language. Four language models are applied to the generated strings, while a trigram model handles the character embeddings. For each language in the sum of probabilities calculated by the four language models, only those languages that produce the highest scores are selected.
To evaluate their approach, the team first had the model translate English into German and used different amounts of target language training data (10,000 and 20,000 utterances, compared to millions of utterances in the complete source language dataset). They then trained transfer models on three different languages (English, German, and Spanish) and transferred them to French (the target language with 10,000 and 20,000 utterances). They claim that the transfer models outperformed the baseline models trained only on the target language, and the models trained using the novel data selection technique performed better than those trained only on the target language.
Google TV integrates Gemini AI for image/video generation, enhances Google Photos, and adds a dedicated YouTube Shorts feed to the homepage.
Read moreAlphabet Q1 paid subscribers reached 350M, fueled by YouTube & Google One. YouTube ad revenue slightly missed at $9.88B amid Premium growth. Total Q1 revenue hit $109.9B, beating forecasts.
Read moreAnthropic releases Claude connectors for Adobe, Blender, and Ableton, enabling AI workflow automation and app control. It also sponsors the Blender Development Foundation.
Read moreApple introduces an App Store model splitting 12-month plans into monthly installments. On iOS 26.4+, it lowers upfront costs without reducing total price, adding billing tracking.
Read moreGoogle launches Aadhaar digital IDs in Wallet for India, enabling secure QR/NFC verification. A phased partner rollout targets key sectors, while passport-based credentials expand globally.
Read moreYouTube tests "Ask YouTube," an AI tool blending text and video for complex queries. Initially for US Premium users, it supports interactive follow-ups, expanding Google's AI search.
Read moreGoogle grants US DoD classified AI access, joining OpenAI and xAI in defense deals amid Anthropic's dispute. Contractual safeguards against surveillance and lethal weapons remain unenforceable.
Read moreMistral AI launches Workflows, a Temporal-powered orchestration platform for secure, Python-driven enterprise AI automation featuring human-in-the-loop approvals.
Read moreCustom subliminal audio maker

AI Social Media Manager

AI Search Visibility

Workout and Calorie Tracker
Stock research platform

AI interview assistant

Scientific figure generator

AI courses with real code