Named entity recognition for novel domains can be challenging in the absence of suitable training materials for machine-learning or lexicons and gazetteers for term look-up. We describe an approach that starts from a small, manually created word list of commodities traded in the nineteenth century, and then uses semantic web techniques to augment the list by an order of magnitude, drawing on data stored in DBpedia. This work was conducted during the Trading Consequences project on text mining and visualisation of historical documents for the study of global trading in the British empire.