Combining Compositional Semantics Modules

点击翻译按钮获取中文摘要

Vaccination aversion is a growing threat in highly mobile populations as a case of measles in San Diego in 2008 demonstrates (http://www.cdc.gov/mmwr/preview/mmwrhtml/mm5708a3.htm). The social and economic cost of avoidable outbreaks is very high and means to improve compliance are under current investigation, including a CIHR-funded project at McGill which studies reasons for vaccine avoidance in blogs in an effort to counteract them through information campaigns (Vaccine Sentiment Ontology http://surveillance.mcgill.ca/projects_theme.php?theme=3). This is only one of the many important issues addressed using text freely available on the internet or on social media. Text processing is an exponential growth area and many join in, not all equipped with essential tools. Anybody can process text: if you accidentally delete half the text, you have processed it. If, per chance, this remaining half has desired information in it, you have summarized successfully. But linguistic sophistication is required if one is to develop general text processing pipelines that work predictably well on a wide variety of texts not trained on, usually leading to highly complex processing environments. Because new ways of looking at more documents than before is the hallmark of exploration and discovery, we are working on just such processing pipelines and hope to strengthen their robustness and their performance into a light-weight, reconfigurable open-source kit of baseline text analysis, that can be harnessed by application builders who do not themselves have this degree of linguistic insight. Our efforts join a growing number of open source modules and we design for interoperability with a variety of already released modules, rather than proposing our own monolithic processing environment.