I use computational linguistic methods to look at the wide, open field of language, and create experiments, tests and prototypes that explore the ways in which we communicate and interact with each other. I am particularly interested in developing work in literature and language, storytelling and conversation, and the fuzzier areas of metaphor, analogy and humour.
Past projects include:
This project was a collaboration with The Poetry Foundation to build poetry recommendation prototype.
What makes a poem similar to another poem? What poetry recommendations would engage a poetry reader? For that matter, what makes up the “genetic code” of a poem, are they attributes such as poetic devices, themes, language use or the like? Automatically detecting and identifying “fuzzy” poetic concepts such as metaphor, allusion and speaker voice is a tricky challenge.
We ran blind tests to gauge how most readers might respond to the idea of poem similarity and found that rather than subject matter playing a strong part in linking two random poems, it was ultimately poetic tone that was a stronger indicator of poem similarity. A recommendation prototype was developed from this work, drawing on 30 identified poetry “genes” from a random corpus of poems.
Conversational Artificial Intelligence
I’ve consulted for tech startups on AI conversation architecture and design to improve the ways information is both delivered and extracted (working with taxonomies, ontologies and knowledge bases) to enhance the user experience, by developing character and language use for chatbot personalities. Chatbots are having a moment, and yet most chatbots built today are simple conversational transactions — essentially wordier versions of nameless and personality-less apps. What goes into the building of character, personality and language use for chatbots when considering a more engaging, persuasive and informative user experience? What is unique about the technical structure of a conversation that can help carry across not just bare bones data from one point to another, but can tell a richer story? I’ve worked with different teams architecting chatbots for use in a one-on-one conversational context, such as for job interviews, online tutoring and content recommendation, as well as a one-to-many context, such as for group/content moderation.
Spelling. Grammar. Literacy. Check.
Not just a hobby horse for pedantic language lovers, spelling, grammar and literacy checkers have come a long way. I’ve worked on developing (as part of working on search engine development) phonetics and phonology for spelling and text to speech (as well as rhyme detection), building bigger dictionaries to take into account new language and unique terms, enhanced grammar rules, as well as literacy level checkers to help users find kid-friendly content, or detect cliched, repetitive or academic content for writing aids. With the advent of the internet and the ubiquity of search and social media, language use has exploded as more neologisms, brand names, meme references have entered the scene. Particularly in English, where orthography is a little more eccentric than in other languages, spell checkers have to keep up the pace with new words and references.