In today's multilingual lexical databases, the majority of the world's languages are under-represented. Beyond a mere issue of resource incompleteness, we show that existing lexical databases have structural limitations that result in a reduced expressivity on culturally-specific words and in mapping them across languages. In particular, the lexical meaning space of dominant languages, such as English, is represented more accurately while linguistically or culturally diverse languages are mapped in an approximate manner. Our paper assesses state-of-the-art multilingual lexical databases and evaluates their strengths and limitations with respect to their expressivity on lexical phenomena of linguistic diversity.

An Italian company has unveiled a novel method of measuring AI progress: analyzing improvements in machine translation. Translated, a provider of translation services, used the approach to predict when we will achieve singularity, a vague concept often defined as the point where machines become smarter than humans. The Rome-based business sets this milestone at the moment when AI provides "a perfect translation." According to the new research, this arrives when machine translation (MT) is better than top human translations. Translated's analysis suggests this will happen before the end of the 2020s.

I've always been interested in computers because of their ability to help people better understand the world around them. Over the last decade, much of the research done at Google has been in pursuit of a similar vision -- to help people better understand the world around them and get things done. We want to build more capable machines that partner with people to accomplish a huge variety of tasks. Analysis and synthesis tasks, like crafting new documents or emails from a few sentences of guidance, or partnering with people to jointly write software together. We want to solve complex mathematical or scientific problems. Transform modalities, or translate the world's information into any language. Diagnose complex diseases, or understand the physical world. We've demonstrated early versions of some of these capabilities in research artifacts, and we've partnered with many teams across Google to ship some of these capabilities in Google products that touch the lives of billions of users. But the most exciting aspects of this journey still lie ahead! With this post, I am kicking off a series in which researchers across Google will highlight some exciting progress we've made in 2022 and present our vision for 2023 and beyond. I will begin with a discussion of language, computer vision, multi-modal models, and generative machine learning models.

In natural language processing (NLP), code-mixing (CM) is a challenging task, especially when the mixed languages include dialects. In Southeast Asian countries such as Singapore, Indonesia, and Malaysia, Hokkien-Mandarin is the most widespread code-mixed language pair among Chinese immigrants, and it is also common in Taiwan. However, dialects such as Hokkien often have a scarcity of resources and the lack of an official writing system, limiting the development of dialect CM research. In this paper, we propose a method to construct a Hokkien-Mandarin CM dataset to mitigate the limitation, overcome the morphological issue under the Sino-Tibetan language family, and offer an efficient Hokkien word segmentation method through a linguistics-based toolkit. Furthermore, we use our proposed dataset and employ transfer learning to train the XLM (cross-lingual language model) for translation tasks. To fit the code-mixing scenario, we adapt XLM slightly. We found that by using linguistic knowledge, rules, and language tags, the model produces good results on CM data translation while maintaining monolingual translation quality.

The use of navigation, online purchases, social media browsing, or streaming services is all impacted by machine learning in one way or another. FREMONT, CA: A new wave of attention is being paid to machine learning, a subset of artificial intelligence. A resurgence in interest in big data is attributed to many factors, including powerful and affordable computational processing, increasing volumes of big data sets, and affordable data storage options. Machine learning is teaching machines to recognize patterns in data and apply them to specific problems. Whenever new data is presented to machine learning models, they adapt independently to make sense of it.

Machine Learning graphs (or models) can be challenging or impossible to train when either devices have limited memory, or models are large. To split the model across devices, learning-based approaches are still popular. While these result in model placements that train fast on data (i.e., low step times), learning-based model-parallelism is time-consuming, taking many hours or days to create a placement plan of operators on devices. We present the Baechi system, the first to adopt an algorithmic approach to the placement problem for running machine learning training graphs on small clusters of memory-constrained devices. We integrate our implementation of Baechi into two popular open-source learning frameworks: TensorFlow and PyTorch. Our experimental results using GPUs show that: (i) Baechi generates placement plans 654 X - 206K X faster than state-of-the-art learning-based approaches, and (ii) Baechi-placed model's step (training) time is comparable to expert placements in PyTorch, and only up to 6.2% worse than expert placements in TensorFlow. We prove mathematically that our two algorithms are within a constant factor of the optimal. Our work shows that compared to learning-based approaches, algorithmic approaches can face different challenges for adaptation to Machine learning systems, but also they offer proven bounds, and significant performance benefits.

English is the international standard of social research, but scholars are increasingly conscious of their responsibility to meet the need for scholarly insight into communication processes globally. This tension is as true in computational methods as any other area, with revolutionary advances in the tools for English language texts leaving most other languages far behind. In this paper, we aim to leverage those very advances to demonstrate that multi-language analysis is currently accessible to all computational scholars. We show that English-trained measures computed after translation to English have adequate-to-excellent accuracy compared to source-language measures computed on original texts. We show this for three major analytics -- sentiment analysis, topic analysis, and word embeddings -- over 16 languages, including Spanish, Chinese, Hindi, and Arabic. We validate this claim by comparing predictions on original language tweets and their backtranslations: double translations from their source language to English and back to the source language. Overall, our results suggest that Google Translate, a simple and widely accessible tool, is effective in preserving semantic content across languages and methods. Modern machine translation can thus help computational scholars make more inclusive and general claims about human communication.

In this paper, we show that the combination of Phrase Pair Injection and Corpus Filtering boosts the performance of Neural Machine Translation (NMT) systems. We extract parallel phrases and sentences from the pseudo-parallel corpus and augment it with the parallel corpus to train the NMT models. With the proposed approach, we observe an improvement in the Machine Translation (MT) system for 3 low-resource language pairs, Hindi-Marathi, English-Marathi, and English-Pashto, and 6 translation directions by up to 2.7 BLEU points, on the FLORES test data. These BLEU score improvements are over the models trained using the whole pseudo-parallel corpus augmented with the parallel corpus.

Abstract--Natural language BERTs are trained with language corpus in a self-supervised manner. An example of generated image descriptions. Given different condition flags, our proposed UCM model is able to generate diverse image descriptions, such as COCO caption, dense caption, and questions. It's clear that the generated contents have different styles. Large scale pretraining has become the dominating approach in various natural language processing tasks. The success of large scale pretraining is due to a large amount of language setting. Although these models can be finetuned to perform training data available everywhere and the self-training algorithm. In this paper, we Second, current common practice in vision language BERT propose a self-training approach that allows to pretrain VL-pretraining uses various image descriptions to train, such as BERTs using unlabeled image data. Those image Self-training is usually done by iterating the following three descriptions have significant differences, making it difficult for steps: 1) training with labeled data, 2) generating pseudo labels an unconditional model to learn to generate adequate pseudo for unlabeled data, 3) mixing the labeled data and unlabeled captions for unlabeled images. However, the has shown its effectiveness in various tasks [4], [5], how to self-training of vision language BERTs is nontrivial due to use it effectively in training vision language BERTs is not yet the following reasons. First, although auto-encoding models studied.

Contrastive learning is widely used for sentence representation learning. Despite this prevalence, most studies have focused exclusively on English and few concern domain adaptation for domain-specific downstream tasks, especially for low-resource languages like Japanese, which are characterized by insufficient target domain data and the lack of a proper training strategy. To overcome this, we propose a novel Japanese sentence representation framework, JCSE (derived from ``Contrastive learning of Sentence Embeddings for Japanese''), that creates training data by generating sentences and synthesizing them with sentences available in a target domain. Specifically, a pre-trained data generator is finetuned to a target domain using our collected corpus. It is then used to generate contradictory sentence pairs that are used in contrastive learning for adapting a Japanese language model to a specific task in the target domain. Another problem of Japanese sentence representation learning is the difficulty of evaluating existing embedding methods due to the lack of benchmark datasets. Thus, we establish a comprehensive Japanese Semantic Textual Similarity (STS) benchmark on which various embedding models are evaluated. Based on this benchmark result, multiple embedding methods are chosen and compared with JCSE on two domain-specific tasks, STS in a clinical domain and information retrieval in an educational domain. The results show that JCSE achieves significant performance improvement surpassing direct transfer and other training strategies. This empirically demonstrates JCSE's effectiveness and practicability for downstream tasks of a low-resource language.