VerbCorner invites citizen scientists to answer fun questions about words and their meanings to eventually help train computers to understand language.
SciStarter is shuffling science into the language department. Explore the science of words with these citizen science projects!
Verb. Noun. Pronoun. Adjective. Adverb. Preposition. Conjunction. Interjection… If you’re anything like me, the sight of sentence diagrams and parts of speech trigger nightmares of grade school English class and number 2 pencils. But, how do we understand what a word means? As useful as dictionaries are, they only provide other words in their definitions. How do we know when to use one word in a sentence and not another? How do we explain a complex idea to children?
In July 2013, Lily Bui introduced us to VerbCorner, a citizen science project investigating the structure of language and, ultimately, the structure of thought. According to Dr. Hartshorne, the director of MIT’s Games With Words, “Scientists still haven’t worked out the exact meaning of most words… It is [similarly] hard to understand how children come to learn the meanings of words, when we don’t fully understand those meanings ourselves… I can tell you as a former translator, we translate words into meaning and then back into words.” With an infinite number of words and sentence structures, where should linguists begin to understand their meanings?
In VerbCorner, the mammoth task of understanding how language is structured is broken down into smaller, simple tasks. Through wild stories, each task focuses on one of seven unique aspects of verb meaning to elucidate the fundamental building blocks scientists think make up language such as how a verb is used and its relationship to a change of state. To date, over 1500 citizen scientists have provided more than 117,000 judgments on the initial 641 chosen verbs and six aspects of meaning, supplying Dr. Hartshorne and colleagues enough preliminary data to begin understanding how we understand words.
Are there patterns to how we use verbs? Do we systematically choose words? Or is our word choice completely random and learned in childhood? Regardless of what language you speak, scientists are learning that how we choose words appears to be similar. Scientists have discovered there is a relationship between grammar and meaning which is systematic, and thereby machine learnable. Previous work by Beth Levin lead to the creation of VerbNet, a database where verbs are categorized based on their meaning and usage. Citizen scientists in VerbCorner are helping Dr. Hartshorne’s team verify that verbs with similar descriptions (such as contact or force) are similarly classified and only work in particular grammatical situations.
According to Dr. Hartshorne, “The most interesting things we are learning are about the building blocks of the mind.” In the initial release of VerbCorner, one task tried to understand the building block ‘change of state’ be it physical, mental, or a location. “If ‘change of state’ really was a core component of how we conceptualize the world, it should have been easy to make a task that got at it. We were unable to make such a task. People found it very hard to keep track of all three types of changes.” Dr. Hartshorne explained. Consequently, the initial task was broken down into three new tasks, each focusing on a different aspect of changing state. Suddenly, citizen scientists were able to complete the task. “This suggests that linguists were wrong about ‘change of state’ being a building block of meaning… Rather the building blocks are probably ‘change of physical state’, ‘change of location’, ‘change of mental state’, and possibly more.”
With these results, VerbCorner achieved its first goal – the analysis of the original 641 verbs and six aspects of meaning. But, there is lots of work still to do – another 400 verbs and four additional tasks have recently been added to the project which ultimately plans to cover all the verbs and components of meaning in VerbNet. Linguists, psychologists, and computer scientists plan to use our evolving understand of language and meaning to develop a deeper understanding of human thought as well as fine tune the artificial intelligence programming society is increasingly reliant upon.
Why not flex your mind over language skills at VerbCorner this afternoon? I’m certain you know more than Siri.
Dr. Melinda T. Hough is a freelance science advocate and communicator. Her previous work has included a Mirzayan Science and Technology Graduate Policy Fellowship at the National Academy of Sciences (2012), co-development of several of the final science policy questions with ScienceDebate.org (2012), consulting on the development of the Seattle Science Festival EXPO day (2012), contributing photographer for JF Derry’s book “Darwin in Scotland” (2010) and outreach projects to numerous to count. Not content to stay stateside, Melinda received a B.S in Microbiology from the University of Washington (2001) before moving to Edinburgh, Scotland where she received a MSc (2002) and PhD (2008) from the University of Edinburgh trying to understand how antibiotics kill bacteria. Naturally curious, it is hard to tear Melinda away from science; but if you can, she might be found exploring, often behind the lens of her Nikon D80, training for two half-marathons, or plotting her next epic adventure.
