American localization specialist Lionbridge Technologies has been employing machine translation tools for many years. Eventually, its customers started asking for multilingual training data. Today, Lionbridge has a separate division entirely dedicated to AI, doing everything from collection of chatbot training data to image annotation, audio transcription and even multilingual content moderation services. To find out more about the work of the division, AI Business talked to Aristotelis Kostopoulos, vice president of product solutions, artificial intelligence at Lionbridge. Q: The AI division at Lionbridge grew out of the machine translation business, but today it does so much more.

At a height of 100m up the amazing A'DAM Tower in central Amsterdam, the altitude wasn't a problem at Global Transformation Summit (GTS) but keeping up with the many shared experiences and fast exchange of ideas was! GTS Amsterdam brought together global brands, connecting international business leaders and senior marketing and localization professionals. What was the common ground? Many insights shared and new contacts made. As content types and volumes continue to increase – the growth of content on the internet doubles every 18 months – brands need to converge content, collaborate internally, and ensure the customer experience is consistent and personal, to stand out from online competition. This means re-imagining how we work – looking to define how multilingual content performs beyond traditional KPIs.

OpenSeq2Seq main goal is to allow researchers to most effectively explore various sequence-to-sequence models. The efficiency is achieved by fully supporting distributed and mixed-precision training. OpenSeq2Seq is built using TensorFlow and provides all the necessary building blocks for training encoder-decoder models for neural machine translation, automatic speech recognition, speech synthesis, and language modeling. Speech-to-text workflow uses some parts of Mozilla DeepSpeech project. Beam search decoder with language model re-scoring implementation (in decoders) is based on Baidu DeepSpeech.

In recent years, natural language processing (NLP) has become a part of our everyday lives. Smartphones now come equipped with NLP-powered voice assistants that interpret and understand human speech in order to provide relevant responses to user queries. NLP also helps translation apps break down communication barriers by analyzing input in one language and transforming it into another language. Even word processors rely on NLP to check the grammar, logic, and syntax of written input. And NLP is now an integral part of customer service; it's used to guide people to the right representative through verbal commands. Yet, few people actually understand how NLP plays a role in making them possible.

Is it a Good Idea to Trust Machine Translations? Although machine translation (MT) has become extremely popular in recent years, it still has a long way to go before it can substitute a human translator. Does it mean that you shouldn t use machine translation?

Training autoregressive models to better predict under the test metric, instead of maximizing the likelihood, has been reported to be beneficial in several use cases but brings additional complications, which prevent wider adoption. In this paper, we follow the learning-to-search approach (Daum\'e III et al., 2009; Leblond et al., 2018) and investigate its several components. First, we propose a way to construct a reference policy based on an alignment between the model output and ground truth. Our reference policy is optimal when applied to the Kendall-tau distance between permutations (appear in the task of word ordering) and helps when working with the METEOR score for machine translation. Second, we observe that the learning-to-search approach benefits from choosing the costs related to the test metrics. Finally, we study the effect of different learning objectives and find that the standard KL loss only learns several high-probability tokens and can be replaced with ranking objectives that target these tokens explicitly.

We investigate the problem of simultaneous machine translation of long-form speech content. We target a continuous speech-to-text scenario, generating translated captions for a live audio feed, such as a lecture or play-by-play commentary. As this scenario allows for revisions to our incremental translations, we adopt a re-translation approach to simultaneous translation, where the source is repeatedly translated from scratch as it grows. This approach naturally exhibits very low latency and high final quality, but at the cost of incremental instability as the output is continuously refined. We experiment with a pipeline of industry-grade speech recognition and translation tools, augmented with simple inference heuristics to improve stability. We use TED Talks as a source of multilingual test data, developing our techniques on English-to-German spoken language translation. Our minimalist approach to simultaneous translation allows us to easily scale our final evaluation to six more target languages, dramatically improving incremental stability for all of them.

We transform reinforcement learning (RL) into a form of supervised learning (SL) by turning traditional RL on its head, calling this Upside Down RL (UDRL). Standard RL predicts rewards, while UDRL instead uses rewards as task-defining inputs, together with representations of time horizons and other computable functions of historic and desired future data. UDRL learns to interpret these input observations as commands, mapping them to actions (or action probabilities) through SL on past (possibly accidental) experience. UDRL generalizes to achieve high rewards or other goals, through input commands such as: get lots of reward within at most so much time! A separate paper [61] on first experiments with UDRL shows that even a pilot version of UDRL can outperform traditional baseline algorithms on certain challenging RL problems. We also introduce a related simple but general approach for teaching a robot to imitate humans. First videotape humans imitating the robot's current behaviors, then let the robot learn through SL to map the videos (as input commands) to these behaviors, then let it generalize and imitate videos of humans executing previously unknown behavior. This Imitate-Imitator concept may actually explain why biological evolution has resulted in parents who imitate the babbling of their babies.

Multi-language speech datasets are scarce and often have small sample sizes in the medical domain. Robust transfer of linguistic features across languages could improve rates of early diagnosis and therapy for speakers of low-resource languages when detecting health conditions from speech. We utilize out-of-domain, unpaired, single-speaker, healthy speech data for training multiple Optimal Transport (OT) domain adaptation systems. We learn mappings from other languages to English and detect aphasia from linguistic characteristics of speech, and show that OT domain adaptation improves aphasia detection over unilingual baselines for French (6% increased F1) and Mandarin (5% increased F1). Further, we show that adding aphasic data to the domain adaptation system significantly increases performance for both French and Mandarin, increasing the F1 scores further (10% and 8% increase in F1 scores for French and Mandarin, respectively, over unilingual baselines).

One Sunday, at one of our weekly salsa sessions, my friend Frank brought along a Danish guest. I knew Frank spoke Danish well, since his mother was Danish, and he, as a child, had lived in Denmark. As for his friend, her English was fluent, as is standard for Scandinavians. However, to my surprise, during the evening's chitchat it emerged that the two friends habitually exchanged emails using Google Translate. Frank would write a message in English, then run it through Google Translate to produce a new text in Danish; conversely, she would write a message in Danish, then let Google Translate anglicize it.