Even after achieving a high level of English proficiency, our accents – along with involuntary code-switching, pronunciation of English words as they are pronounced in our native tongue, and more – may still give us away as EFLs. Accent is the most immediately noticeable feature of EFL speakers. After moving to North America, I was faced with a conflict: Should I preserve my foreign accent and embrace it as part of my identity or try to pass as an American? While the perception that all accents are valid is true, it is also – to some extent – naïve. It not only ignores the desire to assimilate into American culture but also minimizes the impact of implicit biases, which can go as far as labeling people with foreign accents as less competent. Another practical reason to develop a North American accent is to adjust to personal assistants such as Siri and Alexa that often fail to understand foreign accents. At the same time as the world is becoming more progressive and inclusive, language technology sometimes inadvertently pushes us a step back.

Language learning is often regarded as beneficial for developing a higher level of empathy and cultural appreciation. When we connect with people from a different linguistic background than ours, we can catch a glimpse of the rich cultural and linguistic mosaic that makes up our world – and incorporate these insights into our perspective of humanity. We also recognize that there are certain compromises that EFL speakers face when they make English their dominant day-to-day means of communication. One is the loss of proficiency in their native language, which can include forgetting words and code-switching to English; the second is a change in identity as we adapt our sense of self to each language we speak. Examining these crises related to language and identity can help us map out a future for how we want to communicate – and for how language learning and language technologies can help us realize our vision.

Euphemisms, a particular type of idiom especially prevalent in American English, are vague or indirect expressions that often substitute harsh, embarrassing, or unpleasant terms. They are widely used to navigate sensitive topics like death and sex. “Passing away,” for example, has long been an accepted term to describe the act of dying. When euphemisms are in use for the length of time it takes to become lexicalized, they are often replaced with new ones, a phenomenon known as “the euphemism treadmill.” Correctly interpreting and using euphemisms can be difficult for EFL learners – and can lead to misuse since these expressions may rely on relevant cultural knowledge. That is unfortunate, given that euphemisms hold sensitive meanings. Artificial intelligence (AI) writing assistants can now go beyond grammar correction to suggesting edits for more inclusive language, such as replacing “whitelist” with “allow-list” and “landlord” with “property owner.” Such suggestions can help inform EFLs and users from diverse cultures – who carry a different cultural baggage – of unintended bias in their writing. At the same time, these assistants also run the risk of erasing individual and cultural differences.

Apart from the words we speak or write, nonverbal communication – such as tone of voice, facial expressions, eye contact, and gestures – also differs across cultures. For example, travel guides for Italy like to warn against using the 🤌 hand gesture commonly signaling “wait” in many countries, because Italians interpret this gesture as, “What the hell are you saying?” Tech companies are now dipping their toes into analyzing users’ behavior as expressed in nonverbal communication. For example, Zoom is providing business customers with AI tools that can determine users’ emotions during video calls based on facial expressions and tone of voice. Unless companies carefully consider cultural differences, the ramifications could be more algorithmic bias and discrimination.

While what is said can be difficult to understand, what is not said may pose an even bigger challenge. Language is efficient, so often what goes without saying is simply not being said. It is left for the reader or listener to interpret underspecified language and resolve ambiguities, a task that we do seamlessly using our personal experience, knowledge about the world, and commonsense reasoning abilities. In many cases, commonsense knowledge helps EFL learners compensate for low language proficiency. However, what is considered “commonsense” is not always universal. Some commonsense knowledge, especially pertaining to social norms, differs between cultures. Can language technologies help bridge this cultural gap? It depends. Chatbots like ChatGPT seem to have broad knowledge about every possible topic in the world. However, ChatGPT learned about the world from reading all the English text on the web, which is primarily coming from the US, and thus it has a North American lens. In addition, despite being “book smart,” it still lacks basic commonsense reasoning abilities that are employed by us to understand social interactions and navigate the world around us.

When I started working in natural language processing in 2013, I had to explain what work in this area of computer science entails. I told people I was teaching computers to speak English. A decade later, ChatGPT has become a household name, language models are on the news every day, and the field is considered one of the most sought-after. We are experiencing an exciting era in which language technologies are maturing and are increasingly used and deployed.

Automatic translation tools like Google Translate have improved immensely in recent years. Older translation technology selected the sentence that sounded more natural in the target language among multiple prospective word-by-word translations. Conversely, the current tools learn a sentence-level translation function from human translations. Although they are very useful, automatic translation tools don’t work equally well for every pair of languages and every genre and topic. For this reason, automatic translation didn’t yet make second language acquisition obsolete. Mastering English means being able to think in English rather than translating your thoughts from your native language. The language of our thoughts affects our word choice and grammatical constructions, so going through another language might result in incorrect or unnatural sentences. Choosing the right English words involves obstacles such as mispronunciation, malapropism, and inappropriate contexts.

In contrast to the rest of the book, this chapter discusses not what to say in and how to speak English but rather what is not socially acceptable to speak about in North American culture: from offensive language and profanity to sensitive topics such as sex and politics. These taboo subjects differ by culture, and EFL speakers who come from cultures that are more direct might find themselves saying something inappropriate – just as chatbots can sometimes generate offensive content. The developers of chatbots like ChatGPT have programmed filters to prevent them from generating offensive text. Those filters are based on the norms of the developers themselves, most of whom are based in North America, and this can make a chatbot’s refusal to answer some questions seem excessively careful through the lens of other cultures.

Picture, for a moment, enlisting the help of automatic translation when you seek medical attention in a foreign country and need to explain, in no uncertain terms, where you experience pain and in what intensity. I have experienced this in my first year in the US after moving there from Israel. Now consider that I’m not only a user of language technologies but also a researcher working on these technologies. As such, I’m also aware of their limitations. For example, I know that translation systems may translate figurative expressions literally, or that certain inputs can make them generate incorrect “translations” in the form of a religious text.

Although the internet has removed geographical boundaries, transforming the world into a global village, English is still the most dominant language online. New forms of online communication such as emoji and memes have become an integral part of internet language. While it’s tempting to think of such visual communication formats as removing the cultural barriers – after all, emoji appear like a universal alphabet – their interpretation may rely on cultural references.

Models for TAMP problems are complex and challenging to develop. The high-dimensional sensory-motor space and the required integration of metric and symbolic state variables augment the challenges. Machine learning addresses these challenges at both the acting level and the planning level. But ML in robotics faces specific problems: lack of massive data; experiments needed for RL are scarce, very expensive, and difficult to reproduce; realistic sensory-motor simulators remain computationally costly; and expert human input for RL, e.g., for specifying or shaping reward functions or giving advices, is scarce and costly. The functions learned tend to be narrow: transfer of learned behaviors and models across environments and tasks is challenging. This chapter presents approaches for learning reactive sensory-motor skills using deep RL algorithms and methods for learning heuristics to guide a TAMP planner avoiding computation on unlikely feasible movements.

Acting with robots and sensory-motor devices demands the combined capabilities of reasoning both on abstract actions and on concrete motion and manipulation steps. In the robotics literature, this is referred to as "task-aware planning," i.e., planning beyond motion and manipulation. In the AI literature, it is referred to as "combined task and motion planning" (TAMP). This class of TAMP problems, which includes task, motion, and manipulation planning, is the topic of this part. The challenge in TAMP is the integration of symbolic models for task planning with metric models for motion and manipulation. This part introduces the representations and techniques for achieving and controlling motion, navigation, and manipulation actions in robotics. It discusses motion and manipulation planning algorithms, and their integration with task planning in TAMP problems. It covers learning for the combined task and motion-manipulation problems.

Acting, planning and learning are critical cognitive functions for an autonomous actor. Other functions, such as perceiving, monitoring, and goal reasoning, are also needed and can be essential in many applications. This chapter briefly surveys a few such functions and their links to acting, planning, and learning. Section 24.1 discusses perceiving and information gathering: how to model and control perception actions in order to recognize the state of the world and detect objects, events, and activities relevant to the actor while performing its tasks. It discusses semantic mapping and anchoring sensor data to symbols. Section 24.2 is about monitoring, that is, detecting and interpreting discrepancies between predictions and observations, anticipating what needs be monitored, and controlling monitoring actions. Goal reasoning in Section 24.3 is about assessing the relevance of current goals, from observed evolutions, failures, and opportunities to achieve a higher-level assigned mission.