Professor Alessandro Lenci

Professor Alessandro Lenci is a professor of the Department of Philology, Literature and Linguistics of the University of Pisa, Italy. He is an expert in Computational Linguistics, especially Distributional Semantics. He contributed to psycholinguistic phenomena modeling such as verb-argument selections, metonyms processing cost, and lexical complexity, as well as automatic detection of semantic relations like antonymy and hyperonymy. Together with Marco Baroni, he created the Distributional Memory framework, a Distributional Semantic Model widely used in psycholinguistic modeling tasks.

Title:
Smart parrots or baby humans? The knowledge of language in men and machines

Abstract
State-of-the-art natural language processing models have reached an unprecedented ability to “mimic” human linguistic skills, from machine translation to text generation. But how much of this success depends on their having really acquired human-like abilities to learn and use language? One key feature of human cognition, which grounds natural language too, is a particularly sophisticated ability to learn, from a relatively limited exposure to data, knowledge that is general enough to abstract from the input it has been learnt from and can be therefore applied to interpret unseen situations. Understanding to what extent current machines are able to match this human-specific feature lies at the core of the current debate in computational linguistics and AI. However, this entails not only exploring the nature of the generalizations in machines, but also having a clear view on the nature of linguistic generalizations in humans. The former still seem to be far less abstract than human ones, but on the other hand much evidence also suggests that human knowledge of language is less abstract than some theories claim to be. In turn, this raises the question whether we ask machines to master some idealized notion of language, or rather we want them to use language like humans do. This is an old issue that divides linguistic theories too, but it is crucial to understand what the successes and failures of machines can teach us about human language.

 

Professor Christine Ji

Christine Meng Ji is an associate Professor in School of Languages and Cultures at the University of Sydney. She specialises in empirical translation studies, especially data-driven multilingual corpus analyses. She has published on environmental translation, healthcare translation, statistical translation stylistics/authorship attribution, and international multilingual education (statistical translation quality evaluation). Her research has been supported by the British Academy, Japanese Society for the Promotion of Sciences, the Australian Research Council, Economic and Social Research Council of the UK, etc.

Title:
Consumer-Adaptive Health Translation

Abstract:
Around the world, health research, knowledge and information is being translated, consumed across industries, societies, languages, cultures, communities. The way we receive, understand, appraise, apply translated health information has important impact on the development of our health literacy, and everyday health behaviours. Health translation can help reduce widening social and health inequalities, improve the accessibility of health information, healthcare, and medical support to diverse populations, and help develop the health literacy of the global public as a most cost-effective approach to health risk management and disease management. The challenge of adapting these clinical principles from public health to health translation is the lack of theoretical frameworks to inform data-intensive studies of English and multilingual health information. Empirical Translation Studies represents a highly interdisciplinary paradigm in translation studies. ETS is known for the use of data and quantitative methods in the study of translation products and processes. Over the past few decades, much of the scholarly debates in ETS has focused on exploring theoretical hypotheses such as translational universals, norms, or laws. The generalising approach to ETS assumes that the use of translation methods and strategies is conditioned by high-level factors such as the imbalance between source and target cultures, or distinct cognitive patterns between source and target languages, and that these high-level factors can account for the highly diverse expectations of translation among target readers. My current work explores the complexity, diversity, variability of the abilities of the target readers to understand, appraise, utilise translated health materials within the framework of consumer-adaptive health translation (CAHT). The delivery of quality health translations, from nutrition guidelines, clinical instructions, to public health advice and health educational materials, requires the development of robust, reliable empirical research methods to examine, evaluate the usability of translated health resources and information. Clinical guidelines have identified key areas of health information usability which can be adapted for health translation research: information understandability, relevance of health materials to the intended readers, actionability, extensibility (ease to apply health materials in diverse, complex settings). We are working on the development of machine learning algorithms, quantitative models to explore the relations between the profiles/backgrounds of consumers and the types of health readings/materials of varying usability, as well as related research issues such as technology-mediated health translation.