Replacing sensors with text occurrences for commonsense knowledge acquisition.

In this position paper we introduce our ideas for utilizing text corpus for supplementing or replacing sensing devices by word pairs occurrences in a blog corpus and we share results of preliminary experiments. We explain how simple web- mining and information retrieval techniques could help to acquire perceptional knowledge while collecting, processing and combining inputs from actual sensors is still technically difficult, expensive or simply impossible. By utilizing co-occurrences rankings of input nouns and five senses identifying words (mostly adjectives and onomatopoeias) we were able to discover physical characteristics of these nouns. Although the method is in early stage of development and the recall is low, achieved precision of 0.96 seems promising, especially if one needs to store acquired output in knowledge bases as ConceptNet