How to teach Transformers to care about word order

New position encoding scheme improves state-of-the-art performance on several natural-language-processing tasks.

The Transformer is a neural-network architecture that has proven extremely useful for natural-language-processing tasks because it can recognize long-range dependencies. It could, for instance, recognize that in a sentence that includes the word “rented”, the word “flat” is more likely to mean “apartment” than it would be otherwise, even if “rented” is the second word in the sentence and “flat” the 10th.

In its most basic form, the Transformer is indifferent to word order. It can recognize the relationship between “rented” and “flat”, but it doesn’t care which comes first.

Word order, however, can make a big difference to meaning. Consider, for instance, the sentences “We rented a small but clean, well-equipped two-bed flat” and “We rented a small but clean, well-equipped flat-bed truck”.

Position embedding.png
These images map 252 words of an input text sequence (y-axis) against the 512 latent position features identified by two different position-encoding schemes. Lighter colors indicate higher values for features, darker colors lower values. FLOATER (bottom) produces a more regular encoding than an earlier scheme (top), which also learns its position feature set from training data. The vertical lines toward the bottom of the top visualization indicate that the encoding model is simply using the same encoding for input sequences longer than those it saw during training, while FLOATER’s smooth gradation from light to dark demonstrates that its encoding generalizes easily to longer sequences.

Starting with the paper that introduced the Transformer, researchers have proposed a series of position encoders that inject word-order information into the Transformer model. But last week, at the International Conference on Machine Learning, we presented a new position encoder that enables better performance than its predecessors on a range of natural-language-processing (NLP) tasks.

We designed our position encoder so that it can be integrated into existing Transformer models, conferring its benefits to NLP systems that already have been trained extensively on large data sets.

Before the Transformer was introduced in 2017, the most popular architecture for NLP was the long short-term memory, or LSTM. LSTMs process sequenced inputs in order, and each output reflects both the inputs and the outputs that preceded it.

LSTMs are very good at inferring local relationships — a word’s relationships, both syntactic and semantic, with the two or three words that immediately precede it — but they’re not as good at modeling long-range dependencies. That’s where the Transformer excels.

Position encodings are an attempt to achieve the best of both worlds: an awareness of long-range dependencies and a sensitivity to local word order. The ideal position encoding should have three properties:

  1. It should be able to handle sequences of arbitrary length; that is, it shouldn’t be locked in to some maximum sequence length.
  2. It should be learnable from training data; different encodings may work better for different tasks.
  3. It should be efficient; adding position encoding shouldn’t unreasonably inflate the size of the neural model.

Past position encoding schemes have met at best two of these criteria. For instance, the original Transformer paper proposed an encoding based on a family of sinusoidal functions; that encoding remains popular, but it is not learnable.

Our scheme, which we call FLOATER, is the first to meet all three criteria.

The naïve way to encode position would be simply to assign successive numbers to successive words in an input sequence. But this has drawbacks in a machine learning context. If at runtime the model sees a sequence of a length it did not encounter during training, it will be flummoxed about how to proceed.

So most position encoding schemes instead use position vectors, which carry information that can be used to deduce the relative positions of two inputs. If those schemes are fully learnable, however, they tend to inflate the model size; or, to keep model inflation under control, they limit the distances across which relative position can be compared.

Functional approach

Instead of learning to directly compute a position vector from each word in an input sequence, FLOATER learns a function that computes each word’s position vector from that of the word that preceded it.

Learning a general function rather than direct mappings makes FLOATER much more space efficient than other learnable encoding schemes. But a general function can also be applied to any word in a sequence, regardless of its position, so FLOATER is indifferent to sequence length.

Any given manually engineered position function — such as the sinusoidal functions proposed in the original Transformer paper — can be thought of as a special case of the general FLOATER function. So in a pretrained network, we can simply substitute FLOATER for any such function and then fine-tune it on a small set of training data.

Past work on position encoding has shown that re-encoding position information at every layer of a Transformer network improves performance on NLP tasks. If we allowed FLOATER to learn a different function for every layer, the model size would again begin to inflate.

So instead, we learn a single function that is applied at every layer. This results in different position encodings at each layer, however, because the inputs are different. Our experiments indicate that this approach strikes a good balance between model size and performance improvements.

In one set of experiments, we compared our position encoder to its two leading predecessors on four different machine translation tasks and found that it delivered the best results across the board.

In another set of experiments, we added our position encoder to Transformer models that had previously been trained on three different language-understanding and question-answering tasks.

Of 23 distinct tasks, the addition of our position encoder improved performance on 21. The two on which its performance fell slightly short were low-data versions of tasks on which, with larger sets of training data, it improved performance.

Related content

US, VA, Arlington
As a Survey Research Scientist within the Reputation Marketing & Insights team, your primary responsibility will be to help manage our employee communications research program, including a global tracking survey. The work will challenge you to be resourceful, think big while staying connected to the details, translate survey, focus group results, and advanced analytics into strategic direction, and embrace a high degree of change and ambiguity at speed. The scope and scale of what we strive to achieve is immense, but it is also meaningful and energizing. This is an individual contributor role. The right candidate possesses endless curiosity and passion for understanding employee perceptions and what drives them. You have end-to-end experience conducting qualitative research, robust large-scale surveys, campaign measurement, as well as advanced modeling skills to uncover perception drivers. You have proficiency in diving deep into large amounts of data and translating research into actionable insights/recommendations for internal communicators. You are an excellent writer who can effectively communicate data-driven insights and recommendations through written documents, presentations, and other internal communication channels. You are a creative problem-solver who seeks to deeply understand the business/communications so you can tailor research that informs stakeholder decision making and strategic messaging tactics. Key job responsibilities - Design and manage the execution of a global tracking survey focused on employee communications - Develop research to identify and test messages to drive employee perceptions - Use advanced statistical methodologies to better understand the relationship between key internal communications metrics and other related measures of perception (e.g., regression, structural equation modeling, latent growth curve modeling, Shapley analysis, etc.) - Develop causal and semi-causal measurement techniques to evaluate the perception impact of internal communications campaigns - Identify opportunities to simplify existing research processes and operate more nimbly - Engage in strategic discussions with internal partner teams to ensure our research generates actionable and on-point findings About the team This team sits within the CCR organization. Our focus is on conducting research that identifies messaging opportunities and informs communication strategies for Amazon as a brand.
US, CA, Santa Clara
Want to work on frontier, world class, AI-powered experiences for health customers and health providers? The Health Science & Analytics group in Amazon's Health Store & Technology organization is looking for a Senior Manager of Applied Science to lead a group of applied scientists and engineers to work hand in hand with physicians to build the future of AI-powered healthcare experiences. We have an ambitious roadmap which includes scaling recently launched products which are already delighting products and the opportunity to build disruptive, new experiences. This role will be responsible for leading the science and technology teams driving these key innovations on behalf of our customers. Key job responsibilities - Independently manage a team of scientists and engineers to sustainably deliver science driven products. - Define the vision and long-term technical roadmap to achieve multi-year business objectives. - Maintain and raise the science bar of the team’s deliverables and keep the broader Amazon Health Services organization apprised of the latest relevant technical developments in the field. - Work across business, clinical, and technical leaders to disambiguate product requirements and socialize progress towards key goals and deliverables. - Proactively identify risks and shape the technical roadmap in anticipation of industry trends in emerging AI subfields.
US, NY, New York
The Artificial General Intelligence (AGI) team is looking for a passionate, talented, and inventive Senior Applied Scientist to work on pre-training methodologies for Generative Artificial Intelligence (GenAI) models. You will interact closely with our customers and with the academic and research communities. Key job responsibilities Join us to work as an integral part of a team that has experience with GenAI models in this space. We work on these areas: - Scaling laws - Hardware-informed efficient model architecture, low-precision training - Optimization methods, learning objectives, curriculum design - Deep learning theories on efficient hyperparameter search and self-supervised learning - Learning objectives and reinforcement learning methods - Distributed training methods and solutions - AI-assisted research About the team The AGI team has a mission to push the envelope in GenAI with Large Language Models (LLMs) and multimodal systems, in order to provide the best-possible experience for our customers.
US, WA, Seattle
Prime Video is a first-stop entertainment destination offering customers a vast collection of premium programming in one app available across thousands of devices. Prime members can customize their viewing experience and find their favorite movies, series, documentaries, and live sports – including Amazon MGM Studios-produced series and movies; licensed fan favorites; and programming from Prime Video add-on subscriptions such as Apple TV+, Max, Crunchyroll and MGM+. All customers, regardless of whether they have a Prime membership or not, can rent or buy titles via the Prime Video Store, and can enjoy even more content for free with ads. Are you interested in shaping the future of entertainment? Prime Video's technology teams are creating best-in-class digital video experience. As a Prime Video technologist, you’ll have end-to-end ownership of the product, user experience, design, and technology required to deliver state-of-the-art experiences for our customers. You’ll get to work on projects that are fast-paced, challenging, and varied. You’ll also be able to experiment with new possibilities, take risks, and collaborate with remarkable people. We’ll look for you to bring your diverse perspectives, ideas, and skill-sets to make Prime Video even better for our customers. With global opportunities for talented technologists, you can decide where a career Prime Video Tech takes you! Key job responsibilities - Develop ML models for various recommendation & search systems using deep learning, online learning, and optimization methods - Work closely with other scientists, engineers and product managers to expand the depth of our product insights with data, create a variety of experiments to determine the high impact projects to include in planning roadmaps - Stay up-to-date with advancements and the latest modeling techniques in the field - Publish your research findings in top conferences and journals A day in the life We're using advanced approaches such as foundation models to connect information about our videos and customers from a variety of information sources, acquiring and processing data sets on a scale that only a few companies in the world can match. This will enable us to recommend titles effectively, even when we don't have a large behavioral signal (to tackle the cold-start title problem). It will also allow us to find our customer's niche interests, helping them discover groups of titles that they didn't even know existed. We are looking for creative & customer obsessed machine learning scientists who can apply the latest research, state of the art algorithms and ML to build highly scalable page personalization solutions. You'll be a research leader in the space and a hands-on ML practitioner, guiding and collaborating with talented teams of engineers and scientists and senior leaders in the Prime Video organization. You will also have the opportunity to publish your research at internal and external conferences.
US, CA, San Francisco
If you are interested in this position, please apply on Twitch's Career site https://www.twitch.tv/jobs/en/ About Us: Twitch is the world’s biggest live streaming service, with global communities built around gaming, entertainment, music, sports, cooking, and more. It is where thousands of communities come together for whatever, every day. We’re about community, inside and out. You’ll find coworkers who are eager to team up, collaborate, and smash (or elegantly solve) problems together. We’re on a quest to empower live communities, so if this sounds good to you, see what we’re up to on LinkedIn and X, and discover the projects we’re solving on our Blog. Be sure to explore our Interviewing Guide to learn how to ace our interview process. You can work in San Francisco, CA or Seattle, WA. Perks - Medical, Dental, Vision & Disability Insurance - 401(k) - Maternity & Parental Leave - Flexible PTO - Amazon Employee Discount
IN, KA, Bengaluru
AWS Infrastructure Services owns the design, planning, delivery, and operation of all AWS global infrastructure. In other words, we’re the people who keep the cloud running. We support all AWS data centers and all of the servers, storage, networking, power, and cooling equipment that ensure our customers have continual access to the innovation they rely on. We work on the most challenging problems, with thousands of variables impacting the supply chain — and we’re looking for talented people who want to help. You’ll join a diverse team of software, hardware, and network engineers, supply chain specialists, security experts, operations managers, and other vital roles. You’ll collaborate with people across AWS to help us deliver the highest standards for safety and security while providing seemingly infinite capacity at the lowest possible cost for our customers. And you’ll experience an inclusive culture that welcomes bold ideas and empowers you to own them to completion. Do you love problem solving? Are you looking for real world Supply Chain challenges? Do you have a desire to make a major contribution to the future, in the rapid growth environment of Cloud Computing? Amazon Web Services is looking for a highly motivated, Data Scientist to help build scalable, predictive and prescriptive business analytics solutions that supports AWS Supply Chain and Procurement organization. You will be part of the Supply Chain Analytics team working with Global Stakeholders, Data Engineers, Business Intelligence Engineers and Business Analysts to achieve our goals. We are seeking an innovative and technically strong data scientist with a background in optimization, machine learning, and statistical modeling/analysis. This role requires a team member to have strong quantitative modeling skills and the ability to apply optimization/statistical/machine learning methods to complex decision-making problems, with data coming from various data sources. The candidate should have strong communication skills, be able to work closely with stakeholders and translate data-driven findings into actionable insights. The successful candidate will be a self-starter, comfortable with ambiguity, with strong attention to detail and ability to work in a fast-paced and ever-changing environment. Key job responsibilities 1. Demonstrate thorough technical knowledge on feature engineering of massive datasets, effective exploratory data analysis, and model building using industry standard time Series Forecasting techniques like ARIMA, ARIMAX, Holt Winter and formulate ensemble model. 2. Proficiency in both Supervised(Linear/Logistic Regression) and UnSupervised algorithms(k means clustering, Principle Component Analysis, Market Basket analysis). 3. Experience in solving optimization problems like inventory and network optimization . Should have hands on experience in Linear Programming. 4. Work closely with internal stakeholders like the business teams, engineering teams and partner teams and align them with respect to your focus area 5. Detail-oriented and must have an aptitude for solving unstructured problems. You should work in a self-directed environment, own tasks and drive them to completion. 6. Excellent business and communication skills to be able to work with business owners to develop and define key business questions and to build data sets that answer those questions 7. Work with distributed machine learning and statistical algorithms to harness enormous volumes of data at scale to serve our customers About the team Diverse Experiences Amazon values diverse experiences. Even if you do not meet all of the preferred qualifications and skills listed in the job description, we encourage candidates to apply. If your career is just starting, hasn’t followed a traditional path, or includes alternative experiences, don’t let it stop you from applying. Why AWS Amazon Web Services (AWS) is the world’s most comprehensive and broadly adopted cloud platform. We pioneered cloud computing and never stopped innovating — that’s why customers from the most successful startups to Global 500 companies trust our robust suite of products and services to power their businesses. Work/Life Balance We value work-life harmony. Achieving success at work should never come at the expense of sacrifices at home, which is why we strive for flexibility as part of our working culture. When we feel supported in the workplace and at home, there’s nothing we can’t achieve. Inclusive Team Culture AWS values curiosity and connection. Our employee-led and company-sponsored affinity groups promote inclusion and empower our people to take pride in what makes us unique. Our inclusion events foster stronger, more collaborative teams. Our continual innovation is fueled by the bold ideas, fresh perspectives, and passionate voices our teams bring to everything we do. Mentorship and Career Growth We’re continuously raising our performance bar as we strive to become Earth’s Best Employer. That’s why you’ll find endless knowledge-sharing, mentorship and other career-advancing resources here to help you develop into a better-rounded professional.
US, NY, New York
Prime Video is a first-stop entertainment destination offering customers a vast collection of premium programming in one app available across thousands of devices. Prime members can customize their viewing experience and find their favorite movies, series, documentaries, and live sports – including Amazon MGM Studios-produced series and movies; licensed fan favorites; and programming from Prime Video add-on subscriptions such as Apple TV+, Max, Crunchyroll and MGM+. All customers, regardless of whether they have a Prime membership or not, can rent or buy titles via the Prime Video Store, and can enjoy even more content for free with ads. Are you interested in shaping the future of entertainment? Prime Video's technology teams are creating best-in-class digital video experience. As a Prime Video technologist, you’ll have end-to-end ownership of the product, user experience, design, and technology required to deliver state-of-the-art experiences for our customers. You’ll get to work on projects that are fast-paced, challenging, and varied. You’ll also be able to experiment with new possibilities, take risks, and collaborate with remarkable people. We’ll look for you to bring your diverse perspectives, ideas, and skill-sets to make Prime Video even better for our customers. With global opportunities for talented technologists, you can decide where a career Prime Video Tech takes you! We are looking for a self-motivated, passionate and resourceful Applied Scientist to bring diverse perspectives, ideas, and skill-sets to make Prime Video even better for our customers. You will spend your time as a hands-on machine learning practitioner and a research leader. You will play a key role on the team, building and guiding machine learning models from the ground up. At the end of the day, you will have the reward of seeing your contributions benefit millions of Amazon.com customers worldwide. Key job responsibilities - Develop AI solutions for various Prime Video Search systems using Deep learning, GenAI, Reinforcement Learning, and optimization methods; - Work closely with engineers and product managers to design, implement and launch AI solutions end-to-end; - Design and conduct offline and online (A/B) experiments to evaluate proposed solutions based on in-depth data analyses; - Effectively communicate technical and non-technical ideas with teammates and stakeholders; - Stay up-to-date with advancements and the latest modeling techniques in the field; - Publish your research findings in top conferences and journals. About the team Prime Video Search Science team owns science solution to power search experience on various devices, from sourcing, relevance, ranking, to name a few. We work closely with the engineering teams to launch our solutions in production.
US, WA, Bellevue
Are you interested in a unique opportunity to advance the accuracy and efficiency of Artificial General Intelligence (AGI) systems? If so, you're at the right place! As a Quantitative Researcher on our team, you will be working at the intersection of mathematics, computer science, and finance, you will collaborate with a diverse team of engineers in a fast-paced, intellectually challenging environment where innovative thinking is encouraged and rewarded. We operate at Amazon's large scale with the energy of a nimble start-up. If you have a learner's mindset, enjoy solving challenging problems, and value an inclusive team culture, you will thrive in this role, and we hope to hear from you. Key job responsibilities * Conduct statistical analyses on web-scale datasets to develop state-of-the-art multimodal large language models * Conceptualize and develop mathematical models, data sampling and preparation strategies to continuously improve existing algorithms * Identify and utilize data sources to drive innovation and improvements to our LLMs About the team We are passionate engineers and scientists dedicated to pushing the boundaries of innovation. We evaluate and represent the customer perspective through accurate benchmarking.
US, WA, Bellevue
The Artificial General Intelligence (AGI) team is looking for a passionate, talented, and inventive Applied Scientist with a strong deep learning background, to help build industry-leading technology with Large Language Models (LLMs) and multimodal systems. Key job responsibilities As an Applied Scientist with the AGI team, you will work with world-class scientists and engineers to develop novel data, modeling and engineering solutions to support the responsible AI initiatives at AGI. Your work will directly impact our customers in the form of products and services that make use of audio technology. About the team While the rapid advancements in Generative AI have captivated global attention, we see these as just the starting point. Our team is dedicated to pushing the boundaries of what’s possible, leveraging Amazon’s unparalleled ML infrastructure, computing resources, and commitment to responsible AI principles. And Amazon’s leadership principle of customer obsession guides our approach, prioritizing our customers’ needs and preferences each step of the way.
US, CA, Sunnyvale
The Artificial General Intelligence (AGI) team is looking for a highly skilled and experienced Senior Applied Scientist, to lead the development and implementation of algorithms and models for supervised fine-tuning and reinforcement learning through human feedback; with a focus across text, image, and video modalities. As a Senior Applied Scientist, you will play a critical role in driving the development of Generative AI (Gen AI) technologies that can handle Amazon-scale use cases and have a significant impact on our customers' experiences. Key job responsibilities - Collaborate with cross-functional teams of engineers, product managers, and scientists to identify and solve complex problems in GenAI - Design and execute experiments to evaluate the performance of different algorithms and models, and iterate quickly to improve results - Think big about the arc of development of GenAI over a multi-year horizon, and identify new opportunities to apply these technologies to solve real-world problems - Communicate results and insights to both technical and non-technical audiences, including through presentations and written reports - Mentor and guide junior scientists and engineers, and contribute to the overall growth and development of the team