Computational models of word meaning in use
The human capacity for understanding meaning is impressive: in order to understand the image a speaker is trying to evoke with a sentence like "The bat has red eyes", we need to make a host of decisions. These ranging from coarse-grained ones (are we talking about a baseball utensil or a flying mammal?), to fine-grained ones (does the speaker mean that the eyes are bloodshot or that the irises are red?). This apparent ease with which humans process meaning stands in stark contrast with the current capacities of computational systems. Nevertheless, allowing computers to arrive at a deeper understanding of meaning is important if we want to improve appliances like Siri and Alexa, and more generally, the Artificial Intelligence (AI) systems that increasingly form part of our everyday life.
In this project, I will integrate insights from different scientific disciplines to build computational software systems or 'models' that carry out language-related tasks in human-like ways. My work draws inspiration from linguistics, as well as cognitive science, two adjacent fields of research that do not always interact as much with computational linguistics as much as they could. The specific project goals are centred around the question how a computational model can correctly identify the intended meaning of a word in context, both at the coarser-grained (baseball or mammal 'bat') and finer-grained (what is meant with 'red' in 'red eyes') level.
An important aspect of this project is the inclusion of languages that are quite different from English. We do so for two reasons. First, an increasing amount of online text data is written in languages other than English. Second, we want our computational model to be general, or "trainable" on any language, capturing critical underlying principles of ambiguity resolution. To make sure it has this property, we need
to test it on a diverse array of languages as well.
In this project, I will develop, together with students from multiple disciplines, computational models that understand, for instance, that the word 'bright' in 'a bright musician' and 'a bright student' mean approximately the same thing, while 'a bright lantern' means something distinctly different. We will also work to develop models that simulate semantic tasks in a time-dependent way, mirroring the time it takes people to read words in a fragment of text, or the time to decide if something is a real word or not. Exploring how human effort translates proportionally to the amount of effort exhibited by the model, for has great potential to help us build and refine these AI systems. After all, no one can interpret language as well as humans can!