Named Entity Normalization (NEN) is the task of determining entities mentioned in text. NEN is different from Named Entity Recognition (NER) in that NER identifies the occurrence or mention of a named entity in text by does not identify which specific entity it is. The distinction is subtle but critical for extracting chemical information from scientific literature, where understanding which chemical an author is referring to is vital. Perhaps disappointingly, this distinction also means that many recent advances in NER, such as word2vec, LSTM neural networks and conditional random fields, have little relevance in practice. A significant technical challenge with NEN is the need to represent the entity/referent computationally. For many terms there exist suitable ontologies and identifiers, and for chemicals there connection tables and canonical notations such as InChI and SMILES. Hence, the cutting edge of chemical information text mining is not just about the syntax of entities, but also semantics. In this presentation, we describe several entity types at the limits of the current state-of-the-art. These range from handling chemical notations, such as HELM, WLN and molecular line formulae, through more complex entities such as scientific journal names and clinical trials (NCT numbers) to more abstract concepts such as the generic terms alkanes, heterocycles, alloys, solvents or zintl phases. The presentation will also cover chemical nomenclature beyond the traditional IUPAC-like nomenclatures handled by existing name-to-structure software, including inorganic nomenclature, peptides, protein variants and mutants, and even mixtures. In the words of Ludwig Wittgenstein, "The limits of my language means the limits of my world". Increasing what can be handled by chemical information text mining requires not only extending the language it recognizes, but also the world that it can represent.

Current challenges in text-mining for chemical information

2019 Fall National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.