Last year our CEO, Sundar Pichai, announced that Google would invest $1 billion in Africa over the next five years to support a range of initiatives, from improved connectivity to investment in startups, to help boost Africa’s digital transformation.
Africa’s internet economy has the potential to grow to $180 billion by 2025 – 5.2% of the continent’s GDP. To support this growth, over the last year we’ve made progress on helping to enable affordable access and on building products for every African user – helping businesses build their online presence, supporting entrepreneurs spur next-generation technologies, and helping nonprofits to improve lives across the continent.
We’d like to share how we’re delivering on our commitment and partnering with others – policymakers, non-profits, businesses and creators – to make the internet more useful to more people in Africa.
Introducing the first Google Cloud region in Africa
Today we’re announcing our intent to establish a Google Cloud region in South Africa – our first on the continent. South Africa will be joining Google Cloud’s global network of 35 cloud regions and 106 zones worldwide.
The future cloud region in South Africa will bring Google Cloud services closer to our local customers, enabling them to innovate and securely deliver faster, more reliable experiences to their own customers, helping to accelerate their growth. According to research by AlphaBeta Economics for Google Cloud, the South Africa cloud region will contribute more than a cumulative USD 2.1 billion to the country’s GDP, and will support the creation of more than 40,000 jobs by 2030.
Niral Patel, Director for Cloud in Africa announces Google's intention to establish Google's first Cloud region in Africa
Along with the cloud region, we are expanding our network through the Equiano subsea cable and building Dedicated Cloud Interconnect sites in Johannesburg, Cape Town, Lagos and Nairobi. In doing so, we are building full scale Cloud capability for Africa.
Supporting African entrepreneurs
We continue to support African entrepreneurs in growing their businesses and developing their talent. Our recently announced second cohort of the Black Founders Fund builds on the success of last year’s cohort, who raised $97 million in follow-on funding and have employed more than 500 additional staff since they were selected. We’re also continuing our support of African small businesses through the Hustle Academy and Google Business Profiles, and helping job seekers learn skills through Developer Scholarships and Career Certifications.
We’ve also continued to support nonprofits working to improve lives in Africa, with a $40 million cash and in-kind commitment so far. Over the last year this has included:
- $1.5M investment in Career Certifications this year bringing our total Google.org funding to more than $3M since 2021
- A $3 million grant to support AirQo in expanding their work monitoring air quality from Kampala to ten cities in five countries on the continent;
- A team of Googlers who have joined the Tony Elumelu Foundation for 6 months, full-time and pro-bono. The team helped build a new training web and app interface to support the next million African entrepreneurs to grow and fund their businesses.
Across all our initiatives, we continue to work closely with our partners – most recently with the UN to launch the Global Africa Business Initiative (GABI), aimed at accelerating Africa’s economic growth and sustainable development.
Building more helpful products for Africa
We recently announced plans to open the first African product development centre in Nairobi. The centre will develop and build better products for Africans and the world.
Today, we’re launching voice typing support for nine more African languages (isiNdebele, isiXhosa, Kinyarwanda, Northern Sotho, Swati, Sesotho, Tswana, Tshivenda and Xitsonga) in Gboard, the Google keyboard – while 24 new languages are now supported on Google Translate, including Lingala, which is spoken by more than 45 million people across Central Africa.
To make Maps more useful, Street View imagery in Kenya, South Africa, Senegal and Nigeria has had a refresh with nearly 300,000 more kilometres of imagery now helping people virtually explore and navigate neighbourhoods. We’re also extending the service to Rwanda, meaning that Street View is now available in 11 African countries.
In addition to expanding the AI Accra Research Centre earlier this year, theOpen Buildings Project, which mapped buildings across the African continent using machine learning and satellite imagery, is expanding to South and Southeast Asia and is a great example of the AI centre creating solutions for Africa that are useful across the world.
Delivering on our promise
We remain committed to working with our partners in building for Africa together, and helping to unlock the benefits of the digital economy for more people by providing useful products, programmes and investments. We’re doing this by partnering with African organisations, businesses and entrepreneurs. It’s the talent and drive of the individuals in the countries, communities and businesses of Africa that will power Africa’s economic growth.
Machine translation (MT) technology has made significant advances in recent years, as deep learning has been integrated with natural language processing (NLP). Performance on research benchmarks like WMT have soared, and translation services have improved in quality and expanded to include new languages. Nevertheless, while existing translation services cover languages spoken by the majority of people world wide, they only include around 100 languages in total, just over 1% of those actively spoken globally. Moreover, the languages that are currently represented are overwhelmingly European, largely overlooking regions of high linguistic diversity, like Africa and the Americas.
There are two key bottlenecks towards building functioning translation models for the long tail of languages. The first arises from data scarcity; digitized data for many languages is limited and can be difficult to find on the web due to quality issues with Language Identification (LangID) models. The second challenge arises from modeling limitations. MT models usually train on large amounts of parallel (translated) text, but without such data, models must learn to translate from limited amounts of monolingual text, which is a novel area of research. Both of these challenges need to be addressed for translation models to reach sufficient quality.
In “Building Machine Translation Systems for the Next Thousand Languages”, we describe how to build high-quality monolingual datasets for over a thousand languages that do not have translation datasets available and demonstrate how one can use monolingual data alone to train MT models. As part of this effort, we are expanding Google Translate to include 24 under-resourced languages. For these languages, we created monolingual datasets by developing and using specialized neural language identification models combined with novel filtering approaches. The techniques we introduce supplement massively multilingual models with a self supervised task to enable zero-resource translation. Finally, we highlight how native speakers have helped us realize this accomplishment.
Meet the Data
Automatically gathering usable textual data for under-resourced languages is much more difficult than it may seem. Tasks like LangID, which work well for high-resource languages, are unsuccessful for under-resourced languages, and many publicly available datasets crawled from the web often contain more noise than usable data for the languages they attempt to support. In our early attempts to identify under-resourced languages on the web by training a standard Compact Language Detector v3 (CLD3) LangID model, we too found that the dataset was too noisy to be usable.
As an alternative, we trained a Transformer-based, semi-supervised LangID model on over 1000 languages. This model supplements the LangID task with the MAsked Sequence-to-Sequence (MASS) task to better generalize over noisy web data. MASS simply garbles the input by randomly removing sequences of tokens from it, and trains the model to predict these sequences. We applied the Transformer-based model to a dataset that had been filtered with a CLD3 model and trained to recognize clusters of similar languages.
We then applied the open sourced Term Frequency-Inverse Internet Frequency (TF-IIF) filtering to the resulting dataset to find and discard sentences that were actually in related high-resource languages, and developed a variety of language-specific filters to eliminate specific pathologies. The result of this effort was a dataset with monolingual text in over 1000 languages, of which 400 had over 100,000 sentences. We performed human evaluations on samples of 68 of these languages and found that the majority (>70%) reflected high-quality, in-language content.
|The amount of monolingual data per language versus the amount of parallel (translated) data per language. A small number of languages have large amounts of parallel data, but there is a long tail of languages with only monolingual data.|
Meet the Models
Once we had a dataset of monolingual text in over 1000 languages, we then developed a simple yet practical approach for zero-resource translation, i.e., translation for languages with no in-language parallel text and no language-specific translation examples. Rather than limiting our model to an artificial scenario with only monolingual text, we also include all available parallel text data with millions of examples for higher resource languages to enable the model to learn the translation task. Simultaneously, we train the model to learn representations of under-resourced languages directly from monolingual text using the MASS task. In order to solve this task, the model is forced to develop a sophisticated representation of the language in question, developing a complex understanding of how words relate to other words in a sentence.
Relying on the benefits of transfer learning in massively multilingual models, we train a single giant translation model on all available data for over 1000 languages. The model trains on monolingual text for all 1138 languages and on parallel text for a subset of 112 of the higher-resourced languages.
At training time, any input the model sees has a special token indicating which language the output should be in, exactly like the standard formulation for multilingual translation. Our additional innovation is to use the same special tokens for both the monolingual MASS task and the translation task. Therefore, the token translate_to_french may indicate that the source is in English and needs to be translated to French (the translation task), or it may mean that the source is in garbled French and needs to be translated to fluent French (the MASS task). By using the same tags for both tasks, a translate_to_french tag takes on the meaning, “Produce a fluent output in French that is semantically close to the input,” regardless of whether the input is garbled in the same language or in another language entirely. From the model’s perspective, there is not much difference between the two.
Surprisingly, this simple procedure produces high quality zero-shot translations. The BLEU and ChrF scores for the resulting model are in the 10–40 and 20–60 ranges respectively, indicating mid- to high-quality translation. We observed meaningful translations even for highly inflected languages like Quechua and Kalaallisut, despite these languages being linguistically dissimilar to all other languages in the model. However, we only computed these metrics on the small subset of languages with human-translated evaluation sets. In order to understand the quality of translation for the remaining languages, we developed an evaluation metric based on round-trip translation, which allowed us to see that several hundred languages are reaching high translation quality.
To further improve quality, we use the model to generate large amounts of synthetic parallel data, filter the data based on round-trip translation (comparing a sentence translated into another language and back again), and continue training the model on this filtered synthetic data via back-translation and self-training. Finally, we fine-tune the model on a smaller subset of 30 languages and distill it into a model small enough to be served.
|Translation accuracy scores for 638 of the languages supported in our model, using the metric we developed (RTTLangIDChrF), for both the higher-resource supervised languages and the low-resource zero-resource languages.|
Contributions from Native Speakers
Regular communication with native speakers of these languages was critical for our research. We collaborated with over 100 people at Google and other institutions who spoke these languages. Some volunteers helped develop specialized filters to remove out-of-language content overlooked by automatic methods, for instance Hindi mixed with Sanskrit. Others helped with transliterating between different scripts used by the languages, for instance between Meetei Mayek and Bengali, for which sufficient tools didn’t exist; and yet others helped with a gamut of tasks related to evaluation. Native speakers were also key for advising in matters of political sensitivity, like the appropriate name for the language, and the appropriate writing system to use for it. And only native speakers could answer the ultimate question: given the current quality of translation, would it be valuable to the community for Google Translate to support this language?
This advance is an exciting first step toward supporting more language technologies in under-resourced languages. Most importantly, we want to stress that the quality of translations produced by these models still lags far behind that of the higher-resource languages supported by Google Translate. These models are certainly a useful first tool for understanding content in under-resourced languages, but they will make mistakes and exhibit their own biases. As with any ML-driven tool, one should consider the output carefully.
The complete list of new languages added to Google Translate in this update:
We would like to thank Julia Kreutzer, Orhan Firat, Daan van Esch, Aditya Siddhant, Mengmeng Niu, Pallavi Baljekar, Xavier Garcia, Wolfgang Macherey, Theresa Breiner, Vera Axelrod, Jason Riesa, Yuan Cao, Mia Xu Chen, Klaus Macherey, Maxim Krikun, Pidong Wang, Alexander Gutkin, Apurva Shah, Yanping Huang, Zhifeng Chen, Yonghui Wu, and Macduff Hughes for their contributions to the research, engineering, and leadership of this project.
We would also like to extend our deepest gratitude to the following native speakers and members of affected communities, who helped us in a wide variety of ways: Yasser Salah Eddine Bouchareb (Algerian Arabic); Mfoniso Ukwak (Anaang); Bhaskar Borthakur, Kishor Barman, Rasika Saikia, Suraj Bharech (Assamese); Ruben Hilare Quispe (Aymara); Devina Suyanto (Balinese); Allahserix Auguste Tapo, Bakary Diarrassouba, Maimouna Siby (Bambara); Mohammad Jahangir (Baluchi); Subhajit Naskar (Bengali); Animesh Pathak, Ankur Bapna, Anup Mohan, Chaitanya Joshi, Chandan Dubey, Kapil Kumar, Manish Katiyar, Mayank Srivastava, Neeharika, Saumya Pathak, Tanya Sinha, Vikas Singh (Bhojpuri); Bowen Liang, Ellie Chio, Eric Dong, Frank Tang, Jeff Pitman, John Wong, Kenneth Chang, Manish Goregaokar, Mingfei Lau, Ryan Li, Yiwen Luo (Cantonese); Monang Setyawan (Caribbean Javanese); Craig Cornelius (Cherokee); Anton Prokopyev (Chuvash); Rajat Dogra, Sid Dogra (Dogri); Mohamed Kamagate (Dyula); Chris Assigbe, Dan Ameme, Emeafa Doe, Irene Nyavor, Thierry Gnanih, Yvonne Dumor (Ewe); Abdoulaye Barry, Adama Diallo, Fauzia van der Leeuw, Ibrahima Barry (Fulfulde); Isabel Papadimitriou (Greek); Alex Rudnick (Guarani); Mohammad Khdeir (Gulf Arabic); Paul Remollata (Hiligaynon); Ankur Bapna (Hindi); Mfoniso Ukwak (Ibibio); Nze Lawson (Igbo); D.J. Abuy, Miami Cabansay (Ilocano); Archana Koul, Shashwat Razdan, Sujeet Akula (Kashmiri); Jatin Kulkarni, Salil Rajadhyaksha, Sanjeet Hegde Desai, Sharayu Shenoy, Shashank Shanbhag, Shashi Shenoy (Konkani); Ryan Michael, Terrence Taylor (Krio); Bokan Jaff, Medya Ghazizadeh, Roshna Omer Abdulrahman, Saman Vaisipour, Sarchia Khursheed (Kurdish (Sorani));Suphian Tweel (Libyan Arabic); Doudou Kisabaka (Lingala); Colleen Mallahan, John Quinn (Luganda); Cynthia Mboli (Luyia); Abhishek Kumar, Neeraj Mishra, Priyaranjan Jha, Saket Kumar, Snehal Bhilare (Maithili); Lisa Wang (Mandarin Chinese); Cibu Johny (Malayalam); Viresh Ratnakar (Marathi); Abhi Sanoujam, Gautam Thockchom, Pritam Pebam, Sam Chaomai, Shangkar Mayanglambam, Thangjam Hindustani Devi (Meiteilon (Manipuri)); Hala Ajil (Mesopotamian Arabic); Hamdanil Rasyid (Minangkabau); Elizabeth John, Remi Ralte, S Lallienkawl Gangte,Vaiphei Thatsing, Vanlalzami Vanlalzami (Mizo); George Ouais (MSA); Ahmed Kachkach, Hanaa El Azizi (Morrocan Arabic); Ujjwal Rajbhandari (Newari); Ebuka Ufere, Gabriel Fynecontry, Onome Ofoman, Titi Akinsanmi (Nigerian Pidgin); Marwa Khost Jarkas (North Levantine Arabic); Abduselam Shaltu, Ace Patterson, Adel Kassem, Mo Ali, Yonas Hambissa (Oromo); Helvia Taina, Marisol Necochea (Quechua); AbdelKarim Mardini (Saidi Arabic); Ishank Saxena, Manasa Harish, Manish Godara, Mayank Agrawal, Nitin Kashyap, Ranjani Padmanabhan, Ruchi Lohani, Shilpa Jindal, Shreevatsa Rajagopalan, Vaibhav Agarwal, Vinod Krishnan (Sanskrit); Nabil Shahid (Saraiki); Ayanda Mnyakeni (Sesotho, Sepedi); Landis Baker (Seychellois Creole); Taps Matangira (Shona); Ashraf Elsharif (Sudanese Arabic); Sakhile Dlamini (Swati); Hakim Sidahmed (Tamazight); Melvin Johnson (Tamil); Sneha Kudugunta (Telugu); Alexander Tekle, Bserat Ghebremicael, Nami Russom, Naud Ghebre (Tigrinya); Abigail Annkah, Diana Akron, Maame Ofori, Monica Opoku-Geren, Seth Duodu-baah, Yvonne Dumor (Twi); Ousmane Loum (Wolof); and Daniel Virtheim (Yiddish).
Scaling large language models has resulted in significant quality improvements natural language understanding (T5), generation (GPT-3) and multilingual neural machine translation (M4). One common approach to building a larger model is to increase the depth (number of layers) and width (layer dimensionality), simply enlarging existing dimensions of the network. Such dense models take an input sequence (divided into smaller components, called tokens) and pass every token through the full network, activating every layer and parameter. While these large, dense models have achieved state-of-the-art results on multiple natural language processing (NLP) tasks, their training cost increases linearly with model size.
An alternative, and increasingly popular, approach is to build sparsely activated models based on a mixture of experts (MoE) (e.g., GShard-M4 or GLaM), where each token passed to the network follows a separate subnetwork by skipping some of the model parameters. The choice of how to distribute the input tokens to each subnetwork (the “experts”) is determined by small router networks that are trained together with the rest of the network. This allows researchers to increase model size (and hence, performance) without a proportional increase in training cost.
While this is an effective strategy at training time, sending tokens of a long sequence to multiple experts, again makes inference computationally expensive because the experts have to be distributed among a large number of accelerators. For example, serving the 1.2T parameter GLaM model requires 256 TPU-v3 chips. Much like dense models, the number of processors needed to serve an MoE model still scales linearly with respect to the model size, increasing compute requirements while also resulting in significant communication overhead and added engineering complexity.
In “Beyond Distillation: Task-level Mixture-of-Experts for Efficient Inference”, we introduce a method called Task-level Mixture-of-Experts (TaskMoE), that takes advantage of the quality gains of model scaling while still being efficient to serve. Our solution is to train a large multi-task model from which we then extract smaller, stand-alone per-task subnetworks suitable for inference with no loss in model quality and with significantly reduced inference latency. We demonstrate the effectiveness of this method for multilingual neural machine translation (NMT) compared to other mixture of experts models and to models compressed using knowledge distillation.
Training Large Sparsely Activated Models with Task Information
We train a sparsely activated model, where router networks learn to send tokens of each task-specific input to different subnetworks of the model associated with the task of interest. For example, in the case of multilingual NMT, every token of a given language is routed to the same subnetwork. This differs from other recent approaches, such as the sparsely gated mixture of expert models (e.g., TokenMoE), where router networks learn to send different tokens in an input to different subnetworks independent of task.
Inference: Bypassing Distillation by Extracting Subnetworks
A consequence of this difference in training between TaskMoE and models like TokenMoE is in how we approach inference. Because TokenMoE follows the practice of distributing tokens of the same task to many experts at both training and inference time, it is still computationally expensive at inference.
For TaskMoE, we dedicate a smaller subnetwork to a single task identity during training and inference. At inference time, we extract subnetworks by discarding unused experts for each task. TaskMoE and its variants enable us to train a single large multi-task network and then use a separate subnetwork at inference time for each task without using any additional compression methods post-training. We illustrate the process of training a TaskMoE network and then extracting per-task subnetworks for inference below.
|During training, tokens of the same language are routed to the same expert based on language information (either source, target or both) in task-based MoE. Later, during inference we extract subnetworks for each task and discard unused experts.|
To demonstrate this approach, we train models based on the Transformer architecture. Similar to GShard-M4 and GLaM, we replace the feedforward network of every other transformer layer with a Mixture-of-Experts (MoE) layer that consists of multiple identical feedforward networks, the “experts”. For each task, the routing network, trained along with the rest of the model, keeps track of the task identity for all input tokens and chooses a certain number of experts per layer (two in this case) to form the task-specific subnetwork. The baseline dense Transformer model has 143M parameters and 6 layers on both the encoder and decoder. The TaskMoE and TokenMoE that we train are also both 6 layers deep but with 32 experts for every MoE layer and have a total of 533M parameters. We train our models using publicly available WMT datasets, with over 431M sentences across 30 language pairs from different language families and scripts. We point the reader to the full paper for further details.Results
In order to demonstrate the advantage of using TaskMoE at inference time, we compare the throughput, or the number of tokens decoded per second, for TaskMoE, TokenMoE, and a baseline dense model. Once the subnetwork for each task is extracted, TaskMoE is 7x smaller than the 533M parameter TokenMoE model, and it can be served on a single TPUv3 core, instead of 64 cores required for TokenMoE. We see that TaskMoE has a peak throughput twice as high as that of TokenMoE models. In addition, on inspecting the TokenMoE model, we find that 25% of the inference time has been spent in inter-device communication, while virtually no time is spent in communication by TaskMoE.
A popular approach to building a smaller network that still performs well is through knowledge distillation, in which a large teacher model trains a smaller student model with the goal of matching the teacher’s performance. However, this method comes at the cost of additional computation needed to train the student from the teacher. So, we also compare TaskMoE to a baseline TokenMoE model that we compress using knowledge distillation. The compressed TokenMoE model has a size comparable to the per-task subnetwork extracted from TaskMoE.
We find that in addition to being a simpler method that does not need any additional training, TaskMoE improves upon a distilled TokenMoE model by 2.1 BLEU on average across all languages in our multilingual translation model. We note that distillation retains 43% of the performance gains achieved from scaling a dense multilingual model to a TokenMoE, whereas extracting the smaller subnetwork from the TaskMoE model results in no loss of quality.
|BLEU scores (higher is better) comparing a distilled TokenMoE model to the TaskMoE and TokenMoE models with 12 layers (6 on the encoder and 6 on the decoder) and 32 experts. While both approaches improve upon a multilingual dense baseline, TaskMoE improves upon the baseline by 3.1 BLEU on average while distilling from TokenMoE improves upon the baseline by 1.0 BLEU on average.|
The quality improvements often seen with scaling machine learning models has incentivized the research community to work toward advancing scaling technology to enable efficient training of large models. The emerging need to train models capable of generalizing to multiple tasks and modalities only increases the need for scaling models even further. However, the practicality of serving these large models remains a major challenge. Efficiently deploying large models is an important direction of research, and we believe TaskMoE is a promising step towards more inference friendly algorithms that retain the quality gains of scaling.
We would like to first thank our coauthors - Yanping Huang, Ankur Bapna, Maxim Krikun, Dmitry Lepikhin and Minh-Thang Luong. We would also like to thank Wolfgang Macherey, Yuanzhong Xu, Zhifeng Chen and Macduff Richard Hughes for their helpful feedback. Special thanks to the Translate and Brain teams for their useful input and discussions, and the entire GShard development team for their foundational contributions to this project. We would also like to thank Tom Small for creating the animations for the blog post.
Advances on neural machine translation (NMT) have enabled more natural and fluid translations, but they still can reflect the societal biases and stereotypes of the data on which they're trained. As such, it is an ongoing goal at Google to develop innovative techniques to reduce gender bias in machine translation, in alignment with our AI Principles.
One research area has been using context from surrounding sentences or passages to improve gender accuracy. This is a challenge because traditional NMT methods translate sentences individually, but gendered information is not always explicitly stated in each individual sentence. For example, in the following passage in Spanish (a language where subjects aren’t always explicitly mentioned), the first sentence refers explicitly to Marie Curie as the subject, but the second one doesn't explicitly mention the subject. In isolation, this second sentence could refer to a person of any gender. When translating to English, however, a pronoun needs to be picked, and the information needed for an accurate translation is in the first sentence.
|Spanish Text||Translation to English|
|Marie Curie nació en Varsovia. Fue la primera persona en recibir dos premios Nobel en distintas especialidades.||Marie Curie was born in Warsaw. She was the first person to receive two Nobel Prizes in different specialties.|
Advancing translation techniques beyond single sentences requires new metrics for measuring progress and new datasets with the most common context-related errors. Adding to this challenge is the fact that translation errors related to gender (such as picking the correct pronoun or having gender agreement) are particularly sensitive, because they may directly refer to people and how they self identify.
To help facilitate progress against the common challenges on contextual translation (e.g., pronoun drop, gender agreement and accurate possessives), we are releasing the Translated Wikipedia Biographies dataset, which can be used to evaluate the gender bias of translation models. Our intent with this release is to support long-term improvements on ML systems focused on pronouns and gender in translation by providing a benchmark in which translations’ accuracy can be measured pre- and post-model changes.
A Source of Common Translation Errors
Because they are well-written, geographically diverse, contain multiple sentences, and refer to subjects in the third person (and so contain plenty of pronouns), Wikipedia biographies offer a high potential for common translation errors associated with gender. These often occur when articles refer to a person explicitly in early sentences of a paragraph, but there is no explicit mention of the person in later sentences. Some examples:
|Pro-drop in Spanish → English||Marie Curie nació en Varsovia. Recibió el Premio Nobel en 1903 y en 1911.||Marie Curie was born in Warsaw. He received the Nobel Prize in 1903 and in 1911. |
|Neutral possessives in Spanish → English||Marie Curie nació en Varsovia. Su carrera profesional fue desarrollada en Francia.||Marie Curie was born in Warsaw. His professional career was developed in France. |
|Gender agreement in English → German||Marie Curie was born in Warsaw. The distinguished scientist received the Nobel Prize in 1903 and in 1911.||Marie Curie wurde in Varsovia geboren. Der angesehene Wissenschaftler erhielt 1903 und 1911 den Nobelpreis. |
|Gender agreement in English → Spanish||Marie Curie was born in Warsaw. The distinguished scientist received the Nobel Prize in 1903 and in 1911.||Marie Curie nació en Varsovia. El distinguido científico recibió el Premio Nobel en 1903 y en 1911.|
Building the Dataset
The Translated Wikipedia Biographies dataset has been designed to analyze common gender errors in machine translation, such as those illustrated above. Each instance of the dataset represents a person (identified in the biographies as feminine or masculine), a rock band or a sports team (considered genderless). Each instance is represented by a long text translation of 8 to 15 connected sentences referring to that central subject (the person, rock band, or sports team). Articles are written in native English and have been professionally translated to Spanish and German. For Spanish, translations were optimized for pronoun-drop, so the same set could be used to analyze pro-drop (Spanish → English) and gender agreement (English → Spanish).
The dataset was built by selecting a group of instances that has equal representation across geographies and genders. To do this, we extracted biographies from Wikipedia according to occupation, profession, job and/or activity. To ensure an unbiased selection of occupations, we chose nine occupations that represented a range of stereotypical gender associations (either feminine, masculine, or neither) based on Wikipedia statistics. Then, to mitigate any geography-based bias, we divided all these instances based on geographical diversity. For each occupation category, we looked to have one candidate per region (using regions from census.gov as a proxy of geographical diversity). When an instance was associated with a region, we checked that the selected person had a relevant relationship with a country that belongs to a designated region (nationality, place of birth, lived for a big portion of their life, etc.). By using this criteria, the dataset contains entries about individuals from more than 90 countries and all regions of the world.
Although gender is non-binary, we focused on having equal representation of “feminine” and “masculine” entities. It's worth mentioning that because the entities are represented as such on Wikipedia, the set doesn't include individuals that identify as non-binary, as, unfortunately, there are not enough instances currently represented in Wikipedia to accurately reflect the non-binary community. To label each instance as "feminine" or "masculine" we relied on the biographical information from Wikipedia, which contained gender-specific references to the person (she, he, woman, son, father, etc.).
After applying all these filters, we randomly selected an instance for each occupation-region-gender triplet. For each occupation, there are two biographies (one masculine and one feminine), for each of the seven geographic regions.
Finally, we added 12 instances with no gender. We picked rock bands and sports teams because they are usually referred to by non-gendered third person pronouns (such as “it” or singular “they”). The purpose of including these instances is to study over triggering, i.e., when models learn that they are rewarded for producing gender-specific pronouns, leading them to produce these pronouns in cases where they shouldn't.
Results and Applications
This dataset enables a new method of evaluation for gender bias reduction in machine translations (introduced in a previous post). Because each instance refers to a subject with a known gender, we can compute the accuracy of the gender-specific translations that refer to this subject. This computation is easier when translating into English (cases of languages with pro-drop or neutral pronouns) since computation is mainly based on gender-specific pronouns in English. In these cases, the gender datasets have resulted in a 67% reduction in errors on context-aware models vs. previous models. As mentioned before, the neutral entities have allowed us to discover cases of over triggering like the usage of feminine or masculine pronouns to refer to genderless entities. This new dataset also enables new research directions into the performance of different models across types of occupations or geographic regions.
As an example, the dataset allowed us to discover the following improvements in an excerpt of the translated biography of Marie Curie from Spanish.
|Translation result with the previous NMT model.|
|Translation result with the new contextual model.|
This Translated Wikipedia Biographies dataset is the result of our own studies and work on identifying biases associated with gender and machine translation. This set focuses on a specific problem related to gender bias and doesn't aim to cover the whole problem. It's worth mentioning that by releasing this dataset, we don't aim to be prescriptive in determining what's the optimal approach to address gender bias. This contribution aims to foster progress on this challenge across the global research community.
The datasets were built with help from Anja Austermann, Melvin Johnson, Michelle Linch, Mengmeng Niu, Mahima Pushkarna, Apu Shah, Romina Stella, and Kellie Webster.
When my wife and I were flying home from a trip to France a few years ago, our seatmate had just spent a few months exploring the French countryside and staying in small inns. When he learned that I worked on Google Translate, he got excited. He told me Translate’s conversation mode helped him chat with his hosts about family, politics, culture and more. Thanks to Translate, he said, he connected more deeply with people around him while in France.
The passenger I met isn't alone. Google Translate on Android hit one billion installs from the Google Play Store this March, and each one represents a story of people being able to better connect with one another. By understanding 109 languages (and counting!), Translate enables conversation and communication between millions of people which otherwise would have been impossible. And Translate itself has gone through countless changes on the path to one billion installs. Here’s how it has evolved so far.
January 2010: App launches
We released our Android app in January 2010, just over a year after the first commercial Android device was launched. As people started using the new Translate app over the next few years, we added a number of features to improve their experience, including early versions of conversation mode, offline translation and translating handwritten or printed text.
January 2014: 100+ million
Our Android app crossed 100 million installs exactly four years after we first launched it. In 2014, Google acquired QuestVisual, the maker of WordLens. Together with the WordLens team, Translate’s goal was to introduce an advanced visual translation experience in our app. Within eight months, the team delivered the ability to instantly translate text using a phone camera, just as the app reached 200 million installs.
November 2015: 300+ million
As it approached 300 million installs, Translate improved in two major ways. First, revamping Translate's conversation mode enabled two people to converse with each other despite speaking different languages, helping people in their everyday lives, as featured in the video From Syria to Canada.
Second, Google Translate's rollout of Neural Machine Translation, well underway when the app reached 500 million installs, greatly improved the fluency of our translations across text, speech and visual translation. As the installs continued to grow, we compressed those advanced models down to a size that can run on a phone. Offline translations made these high-quality translations available to anyone even when there is no network or connectivity is poor.
June 2019: 750+ millionAt 750 million installs, four years after Word Lens integrated into Translate, we launched a major revamp of the instant camera translation experience. This upgrade allowed us to visually translate 88 languages into more than 100 languages.
February 2020: 850+ million
Transcribe, our long-form speech translation feature, launched when we reached 850 million installs. We partnered with the Pixel Buds team to offer streaming speech translations on top of our Transcribe feature, for more natural conversations between people speaking different languages. During this time, we improved the accuracy and increased the number of supported languages for offline translation.
March 2021: 1 billion — and beyond
Aside from these features, our engineering team has spent countless hours on bringing our users a simple-to-use experience on a stable app, keeping up with platform needs and rigorously testing changes before they launch. As we celebrate this milestone and all our users whose experiences make the work meaningful, we also celebrate our engineers who build with care, our designers who fret over every pixel and our product team who bring focus.
Our mission is to enable everyone, everywhere to understand the world and express themselves across languages. Looking beyond one billion installs, we’re looking forward to continually improving translation quality and user experiences, supporting more languages and helping everyone communicate, every day.
The transcription feature in the Google Translate app may be used to create a live, translated transcription for events like meetings and speeches, or simply for a story at the dinner table in a language you don’t understand. In such settings, it is useful for the translated text to be displayed promptly to help keep the reader engaged and in the moment.
However, with early versions of this feature the translated text suffered from multiple real-time revisions, which can be distracting. This was because of the non-monotonic relationship between the source and the translated text, in which words at the end of the source sentence can influence words at the beginning of the translation.
|Transcribe (old) — Left: Source transcript as it arrives from speech recognition. Right: Translation that is displayed to the user. The frequent corrections made to the translation interfere with the reading experience.|
Today, we are excited to describe some of the technology behind a recently released update to the transcribe feature in the Google Translate app that significantly reduces translation revisions and improves the user experience. The research enabling this is presented in two papers. The first formulates an evaluation framework tailored to live translation and develops methods to reduce instability. The second demonstrates that these methods do very well compared to alternatives, while still retaining the simplicity of the original approach. The resulting model is much more stable and provides a noticeably improved reading experience within Google Translate.
|Transcribe (new) — Left: Source transcript as it arrives from speech recognition. Right: Translation that is displayed to the user. At the cost of a small delay, the translation now rarely needs to be corrected.|
Evaluating Live Translation
Before attempting to make any improvements, it was important to first understand and quantifiably measure the different aspects of the user experience, with the goal of maximizing quality while minimizing latency and instability. In “Re-translation Strategies For Long Form, Simultaneous, Spoken Language Translation”, we developed an evaluation framework for live-translation that has since guided our research and engineering efforts. This work presents a performance measure using the following metrics:
- Erasure: Measures the additional reading burden on the user due to instability. It is the number of words that are erased and replaced for every word in the final translation.
- Lag: Measures the average time that has passed between when a user utters a word and when the word’s translation displayed on the screen becomes stable. Requiring stability avoids rewarding systems that can only manage to be fast due to frequent corrections.
- BLEU score: Measures the quality of the final translation. Quality differences in intermediate translations are captured by a combination of all metrics.
It is important to recognize the inherent trade-offs between these different aspects of quality. Transcribe enables live-translation by stacking machine translation on top of real-time automatic speech recognition. For each update to the recognized transcript, a fresh translation is generated in real time; several updates can occur each second. This approach placed Transcribe at one extreme of the 3 dimensional quality framework: it exhibited minimal lag and the best quality, but also had high erasure. Understanding this allowed us to work towards finding a better balance.
One straightforward solution to reduce erasure is to decrease the frequency with which translations are updated. Along this line, “streaming translation” models (for example, STACL and MILk) intelligently learn to recognize when sufficient source information has been received to extend the translation safely, so the translation never needs to be changed. In doing so, streaming translation models are able to achieve zero erasure.
The downside with such streaming translation models is that they once again take an extreme position: zero erasure necessitates sacrificing BLEU and lag. Rather than eliminating erasure altogether, a small budget for occasional instability may allow better BLEU and lag. More importantly, streaming translation would require retraining and maintenance of specialized models specifically for live-translation. This precludes the use of streaming translation in some cases, because keeping a lean pipeline is an important consideration for a product like Google Translate that supports 100+ languages.
In our second paper, “Re-translation versus Streaming for Simultaneous Translation”, we show that our original “re-translation” approach to live-translation can be fine-tuned to reduce erasure and achieve a more favorable erasure/lag/BLEU trade-off. Without training any specialized models, we applied a pair of inference-time heuristics to the original machine translation models — masking and biasing.
The end of an on-going translation tends to flicker because it is more likely to have dependencies on source words that have yet to arrive. We reduce this by truncating some number of words from the translation until the end of the source sentence has been observed. This masking process thus trades latency for stability, without affecting quality. This is very similar to delay-based strategies used in streaming methods such as Wait-k, but applied only during inference and not during training.
Neural machine translation often “see-saws” between equally good translations, causing unnecessary erasure. We improve stability by biasing the output towards what we have already shown the user. On top of reducing erasure, biasing also tends to reduce lag by stabilizing translations earlier. Biasing interacts nicely with masking, as masking words that are likely to be unstable also prevents the model from biasing toward them. However, this process does need to be tuned carefully, as a high bias, along with insufficient masking, may have a negative impact on quality.
The combination of masking and biasing, produces a re-translation system with high quality and low latency, while virtually eliminating erasure. The table below shows how the metrics react to the heuristics we introduced and how they compare to the other systems discussed above. The graph demonstrates that even with a very small erasure budget, re-translation surpasses zero-flicker streaming translation systems (MILk and Wait-k) trained specifically for live-translation.
|+ Stabilization (new)||20.2||4.1||0.1|
|Evaluation of re-translation on IWSLT test 2018 Engish-German (TED talks) with and without the inference-time stabilization heuristics of masking and biasing. Stabilization drastically reduces erasure. Translation quality, measured in BLEU, is very slightly impacted due to biasing. Despite masking, the effective lag remains the same because the translation stabilizes sooner.|
|Comparison of re-translation with stabilization and specialized streaming models (Wait-k and MILk) on WMT 14 English-German. The BLEU-lag trade-off curve for re-translation is obtained via different combinations of bias and masking while maintaining an erasure budget of less than 2 words erased for every 10 generated. Re-translation offers better BLEU / lag trade-offs than streaming models which cannot make corrections and require specialized training for each trade-off point.|
The solution outlined above returns a decent translation very quickly, while allowing it to be revised as more of the source sentence is spoken. The simple structure of re-translation enables the application of our best speech and translation models with minimal effort. However, reducing erasure is just one part of the story — we are also looking forward to improving the overall speech translation experience through new technology that can reduce lag when the translation is spoken, or that can enable better transcriptions when multiple people are speaking.
Thanks to Te I, Dirk Padfield, Pallavi Baljekar, Goerge Foster, Wolfgang Macherey, John Richardson, Kuang-Che Lee, Byran Lin, Jeff Pittman, Sami Iqram, Mengmeng Niu, Macduff Hughes, Chris Kau, Nathan Bain.
Advances in machine learning (ML) have driven improvements to automated translation, including the GNMT neural translation model introduced in Translate in 2016, that have enabled great improvements to the quality of translation for over 100 languages. Nevertheless, state-of-the-art systems lag significantly behind human performance in all but the most specific translation tasks. And while the research community has developed techniques that are successful for high-resource languages like Spanish and German, for which there exist copious amounts of training data, performance on low-resource languages, like Yoruba or Malayalam, still leaves much to be desired. Many techniques have demonstrated significant gains for low-resource languages in controlled research settings (e.g., the WMT Evaluation Campaign), however these results on smaller, publicly available datasets may not easily transition to large, web-crawled datasets.
In this post, we share some recent progress we have made in translation quality for supported languages, especially for those that are low-resource, by synthesizing and expanding a variety of recent advances, and demonstrate how they can be applied at scale to noisy, web-mined data. These techniques span improvements to model architecture and training, improved treatment of noise in datasets, increased multilingual transfer learning through M4 modeling, and use of monolingual data. The quality improvements, which averaged +5 BLEU score over all 100+ languages, are visualized below.
|BLEU score of Google Translate models since shortly after its inception in 2006. The improvements since the implementation of the new techniques over the last year are highlighted at the end of the animation.|
Hybrid Model Architecture: Four years ago we introduced the RNN-based GNMT model, which yielded large quality improvements and enabled Translate to cover many more languages. Following our work decoupling different aspects of model performance, we have replaced the original GNMT system, instead training models with a transformer encoder and an RNN decoder, implemented in Lingvo (a TensorFlow framework). Transformer models have been demonstrated to be generally more effective at machine translation than RNN models, but our work suggested that most of these quality gains were from the transformer encoder, and that the transformer decoder was not significantly better than the RNN decoder. Since the RNN decoder is much faster at inference time, we applied a variety of optimizations before coupling it with the transformer encoder. The resulting hybrid models are higher-quality, more stable in training, and exhibit lower latency.
Web Crawl: Neural Machine Translation (NMT) models are trained using examples of translated sentences and documents, which are typically collected from the public web. Compared to phrase-based machine translation, NMT has been found to be more sensitive to data quality. As such, we replaced the previous data collection system with a new data miner that focuses more on precision than recall, which allows the collection of higher quality training data from the public web. Additionally, we switched the web crawler from a dictionary-based model to an embedding based model for 14 large language pairs, which increased the number of sentences collected by an average of 29 percent, without loss of precision.
Modeling Data Noise: Data with significant noise is not only redundant but also lowers the quality of models trained on it. In order to address data noise, we used our results on denoising NMT training to assign a score to every training example using preliminary models trained on noisy data and fine-tuned on clean data. We then treat training as a curriculum learning problem — the models start out training on all data, and then gradually train on smaller and cleaner subsets.
Advances That Benefited Low-Resource Languages in Particular
Back-Translation: Widely adopted in state-of-the-art machine translation systems, back-translation is especially helpful for low-resource languages, where parallel data is scarce. This technique augments parallel training data (where each sentence in one language is paired with its translation) with synthetic parallel data, where the sentences in one language are written by a human, but their translations have been generated by a neural translation model. By incorporating back-translation into Google Translate, we can make use of the more abundant monolingual text data for low-resource languages on the web for training our models. This is especially helpful in increasing fluency of model output, which is an area in which low-resource translation models underperform.
M4 Modeling: A technique that has been especially helpful for low-resource languages has been M4, which uses a single, giant model to translate between all languages and English. This allows for transfer learning at a massive scale. As an example, a lower-resource language like Yiddish has the benefit of co-training with a wide array of other related Germanic languages (e.g., German, Dutch, Danish, etc.), as well as almost a hundred other languages that may not share a known linguistic connection, but may provide useful signal to the model.
Judging Translation Quality
A popular metric for automatic quality evaluation of machine translation systems is the BLEU score, which is based on the similarity between a system’s translation and reference translations that were generated by people. With these latest updates, we see an average BLEU gain of +5 points over the previous GNMT models, with the 50 lowest-resource languages seeing an average gain of +7 BLEU. This improvement is comparable to the gain observed four years ago when transitioning from phrase-based translation to NMT.
Although BLEU score is a well-known approximate measure, it is known to have various pitfalls for systems that are already high-quality. For instance, several works have demonstrated how the BLEU score can be biased by translationese effects on the source side or target side, a phenomenon where translated text can sound awkward, containing attributes (like word order) from the source language. For this reason, we performed human side-by-side evaluations on all new models, which confirmed the gains in BLEU.
In addition to general quality improvements, the new models show increased robustness to machine translation hallucination, a phenomenon in which models produce strange “translations” when given nonsense input. This is a common problem for models that have been trained on small amounts of data, and affects many low-resource languages. For example, when given the string of Telugu characters “ష ష ష ష ష ష ష ష ష ష ష ష ష ష ష”, the old model produced the nonsensical output “Shenzhen Shenzhen Shaw International Airport (SSH)”, seemingly trying to make sense of the sounds, whereas the new model correctly learns to transliterate this as “Sh sh sh sh sh sh sh sh sh sh sh sh sh sh sh sh sh”.
Although these are impressive strides forward for a machine, one must remember that, especially for low-resource languages, automatic translation quality is far from perfect. These models still fall prey to typical machine translation errors, including poor performance on particular genres of subject matter (“domains”), conflating different dialects of a language, producing overly literal translations, and poor performance on informal and spoken language.
Nonetheless, with this update, we are proud to provide automatic translations that are relatively coherent, even for the lowest-resource of the 108 supported languages. We are grateful for the research that has enabled this from the active community of machine translation researchers in academia and industry.
This effort is built on contributions from Tao Yu, Ali Dabirmoghaddam, Klaus Macherey, Pidong Wang, Ye Tian, Jeff Klingner, Jumpei Takeuchi, Yuichiro Sawai, Hideto Kazawa, Apu Shah, Manisha Jain, Keith Stevens, Fangxiaoyu Feng, Chao Tian, John Richardson, Rajat Tibrewal, Orhan Firat, Mia Chen, Ankur Bapna, Naveen Arivazhagan, Dmitry Lepikhin, Wei Wang, Wolfgang Macherey, Katrin Tomanek, Qin Gao, Mengmeng Niu, and Macduff Hughes.
Machine learning (ML) models for language translation can be skewed by societal biases reflected in their training data. One such example, gender bias, often becomes more apparent when translating between a gender-specific language and one that is less-so. For instance, Google Translate historically translated the Turkish equivalent of “He/she is a doctor” into the masculine form, and the Turkish equivalent of “He/she is a nurse” into the feminine form.
In line with Google’s AI Principles, which emphasizes the importance to avoid creating or reinforcing unfair biases, in December 2018 we announced gender-specific translations. This feature in Google Translate provides options for both feminine and masculine translations when translating queries that are gender-neutral in the source language. For this work, we developed a three-step approach, which involved detecting gender-neutral queries, generating gender-specific translations and checking for accuracy. We used this approach to enable gender-specific translations for phrases and sentences in Turkish-to-English and have now expanded this approach for English-to-Spanish translations, the most popular language-pair in Google Translate.
|Left: Early example of the translation of a gender neutral English phrase to a gender-specific Spanish counterpart. In this case, only a biased example is given. Right: The new Translate provides both a feminine and a masculine translation option.|
Today, along with the release of the new English-to-Spanish gender-specific translations, we announce an improved approach that uses a dramatically different paradigm to address gender bias by rewriting or post-editing the initial translation. This approach is more scalable, especially when translating from gender-neutral languages to English, since it does not require a gender-neutrality detector. Using this approach we have expanded gender-specific translations to include Finnish, Hungarian, and Persian-to-English. We have also replaced the previous Turkish-to-English system using the new rewriting-based method.
Rewriting-Based Gender-Specific Translation
The first step in the rewriting-based method is to generate the initial translation. The translation is then reviewed to identify instances where a gender-neutral source phrase yielded a gender-specific translation. If that is the case, we apply a sentence-level rewriter to generate an alternative gendered translation. Finally, both the initial and the rewritten translations are reviewed to ensure that the only difference is the gender.
|Top: The original approach. Bottom: The new rewriting-based approach.|
Building a rewriter involved generating millions of training examples composed of pairs of phrases, each of which included both masculine and feminine translations. Because such data was not readily available, we generated a new dataset for this purpose. Starting with a large monolingual dataset, we programmatically generated candidate rewrites by swapping gendered pronouns from masculine to feminine, or vice versa. Since there can be multiple valid candidates, depending on the context — for example the feminine pronoun “her” can map to either “him” or “his” and the masculine pronoun “his” can map to “her” or “hers” — a mechanism was needed for choosing the correct one. To resolve this tie, one can either use a syntactic parser or a language model. Because a syntactic parsing model would require training with labeled datasets in each language, it is less scalable than a language model, which can learn in an unsupervised fashion. So, we select the best candidate using an in-house language model trained on millions of English sentences.
|This table demonstrates the data generation process. We start with the input, generate candidates and finally break the tie using a language model.|
We also devised a new method of evaluation, named bias reduction, which measures the relative reduction of bias between the new translation system and the existing system. Here “bias” is defined as making a gender choice in the translation that is unspecified in the source. For example, if the current system is biased 90% of the time and the new system is biased 45% of the time, this results in a 50% relative bias reduction. Using this metric, the new approach results in a bias reduction of ≥90% for translations from Hungarian, Finnish and Persian-to-English. The bias reduction of the existing Turkish-to-English system improved from 60% to 95% with the new approach. Our system triggers gender-specific translations with an average precision of 97% (i.e., when we decide to show gender-specific translations we’re right 97% of the time).
This effort has been successful thanks to the hard work of many people, including, but not limited to the following (in alphabetical order of last name): Anja Austermann, Jennifer Choi, Hossein Emami, Rick Genter, Megan Hancock, Mikio Hirabayashi, Macduff Hughes, Tolga Kayadelen, Mira Keskinen, Michelle Linch, Klaus Macherey, Gergely Morvay, Tetsuji Nakagawa, Thom Nelson, Mengmeng Niu, Jennimaria Palomaki, Alex Rudnick, Apu Shah, Jason Smith, Romina Stella, Vilis Urban, Colin Young, Angie Whitnah, Pendar Yousefi, Tao Yu
Recently, I was at my friend’s family gathering, where her grandmother told a story from her childhood. I could see that she was excited to share it with everyone but there was a problem—she told the story in Spanish, a language that I don’t understand. I pulled out Google Translate to transcribe the speech as it was happening. As she was telling the story, the English translation appeared on my phone so that I could follow along—it fostered a moment of understanding that would have otherwise been lost. And now anyone can do this—starting today, you can use the Google Translate Android app to transcribe foreign language speech as it’s happening.
Transcribe will be rolling out in the next few days with support for any combination of the following eight languages: English, French, German, Hindi, Portuguese, Russian, Spanish and Thai.
To try the transcribe feature, go to your Translate app on Android, and make sure you have the latest updates from the Play store. Tap on the “Transcribe” icon from the home screen and select the source and target languages from the language dropdown at the top. You can pause or restart transcription by tapping on the mic icon. You also can see the original transcript, change the text size or choose a dark theme in the settings menu.
We’ll continue to make speech translations available in a variety of situations. Right now, the transcribe feature will work best in a quiet environment with one person speaking at a time. In other situations, the app will still do its best to provide the gist of what's being said. Conversation mode in the app will continue to help you to have a back and forth translated conversation with someone.
Try it out and give us feedback on how we can be better.