An Instance Level Approach for Shallow Semantic Parsing in Scientific Procedural Text

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 3010–3017 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 3010 An Instance Level Approach for Shallow Semantic Parsing in Scientific Procedural Text Daivik Swarup, Ahsaas Bajaj, Sheshera Mysore Tim O’Gorman, Rajarshi Das, Andrew McCallum {dswarupogguv, abajaj, smysore togorman, rajarshi, mccallum }@cs.umass.edu Abstract In specific domains, such as procedural scien- tific text, human labeled data for shallow se- mantic parsing is especially limited and expen- sive to create. Fortunately, such specific do- mains often use rather formulaic writing, such that the different ways of expressing relations in a small number of grammatically similar labeled sentences may provide high coverage of semantic structures in the corpus, through an appropriately rich similarity metric. In light of this opportunity, this paper explores an instance-based approach to the relation predic- tion sub-task within shallow semantic parsing, in which semantic labels from structurally sim- ilar sentences in the training set are copied to test sentences. Candidate similar sentences are retrieved using SciBERT embeddings. For la- bels where it is possible to copy from a simi- lar sentence we employ an instance level copy network, when this is not possible, a globally shared parametric model is employed. Exper- iments show our approach outperforms both baseline and prior methods by 0.75 to 3 F1 ab- solute in the Wet Lab Protocol Corpus and 1 F1 absolute in the Materials Science Procedu- ral Text Corpus. 1 Introduction Being able to represent natural language descrip- tions of scientific experiments in a structured form promises to allow tackling a range of challenges from automating biomedical experimental proto- cols (Kulkarni et al., 2018) to gaining materials science insight by large scale mining of the lit- erature (Mysore et al., 2019). To facilitate these applications, recent work has created datasets an- notated with sentence level semantic structure for procedural scientific text from experimental biol- ogy (Kulkarni et al., 2018) and materials science (Mysore et al., 2019). However, these corpora, the Wet Lab Protocols corpus (WLP) and the Materials Query: “Centrifuge the sample at 14,000xg for 5 minutes.” Neighbor: “Centrifuge supernatant at 12,000xg for 10 minutes.” Query: “Add 700µl 70% ethanol to the tube and invert several times to wash the DNA pellet.” Neighbor: “Add 200µl 70% ethanol and invert the tube twice to wash the pellet.” Figure 1: Example sentences from the WLP corpus, and their nearest neighbours based on sentence repre- sentations obtained from SCIBERT. Science Procedural Text (MSPT) corpus remain small. This motivates approaches to parsing that are likely to generalize given limited labelled data. We propose an instance-based edge-factored ap- proach for the relation prediction sub-problem of shallow semantic parsing. To predict a possible re- lation between two entities, our approach retrieves a set of sentences similar to the target sentence, and learns to copy relations in those sentences to the target sentence (Figure 1 shows some examples). However, using only a nearest-neighbours ap- proach over similar sentences poses a coverage problem, as some edge labels may have …

Content truncated to 500 words