Argumentation Mining in Scientific Discourse Nancy L. Green University of North Carolina Greensboro, Greensboro, NC, USA nlgreen@uncg.edu Abstract audience, as well as constraints of the underlying argumentation scheme [Green, 2010]. The dominant approach to argumentation mining has been to treat argumentation Although human-level understanding of natural scheme detection as a machine learning language text is currently beyond the state of the art, problem based upon superficial text features, we contend that an inference-based approach is and to treat the relationships between feasible for applications requiring a deeper analysis arguments as support or attack. However, of argumentation. In [Green, 2016] we proposed an applications such as accurately representing approach to mining individual arguments in and summarizing argumentation in scientific biomedical research articles using argumentation research articles require a deeper schemes implemented as logic programs. The understanding of the text and a richer model schemes are formulated in terms of semantic of relationships between arguments. This predicates that could be obtained from a text by use paper presents a semantic rule-based of BioNLP (biomedical/biological natural language approach to extracting individual arguments, processing) tools. This semantic approach to mining and demonstrates the need for a richer avoids the various problems faced by purely feature- model of inter-argument relationships in based approaches, e.g., that argument components biomedical/biological research articles. may be conveyed through non-contiguous or overlapping text segments of varying granularity, the sparcity of discourse cues marking argument 1 Introduction components, and the occurrence of enthymemes. The dominant approach to argumentation (or In this paper, we build on our previous proposal by argument) mining [e.g., Green et al., 2014; Cardie et considering the role of discourse structure in mining al., 2015; Reed et al., 2016] has been to treat it as a argumentation in scientific texts. In section 2 we machine learning problem based upon superficial text summarize our previous proposal to mining features, enabling researchers to adopt methods that individual arguments. In section 3 we discuss the have been applied successfully to other natural relationship of the individual arguments to other language processing tasks. This approach has been aspects of discourse structure, and how the arguments useful for applications such as identifying reasons are related to each other, i.e., the argumentation given for opinions in social media, or automatic structure of the discourse. assessment of student essay quality. However some applications, such as accurately summarizing 2 Mining Arguments argumentation in scientific research articles, require a This section summarizes our proposed approach to deeper understanding of the text. mining individual arguments described in [Green, There are a number of problems with mining 2016], using argumentation schemes implemented as arguments in scientific documents at the text level logic programs written in Prolog [Bratko, 2001]. rather than at the semantic level [Green, 2015a; Argumentation schemes are abstract descriptions of 2015b]. Argument components may not occur in acceptable, possibly defeasible, arguments used in contiguity. In fact, the content of an argument may be conversation as well as in formal genres such as legal widely separated or the content of two arguments and scientific text [Walton et al., 2008]. To provide may be interleaved at the text level. Furthermore examples of argumentation schemes in open-access scientific text often contains enthymemes, i.e. text, we analyzed arguments in the Results section of arguments with implicit premises or an implicit a biomedical research article [van de Leemput et al., conclusion. Interpretation of enthymemes may 2007] in the CRAFT corpus [CRAFT]. The CRAFT require use of the preceding discourse context corpus has been annotated by other researchers for (including inferred conclusions of other arguments), purposes of biomedical text mining [Verspoor et al., presumed shared knowledge of the author and 2012; Bada et al., 2012], but not for argument mining. The seven argumentation schemes presented 18th Workshop on Computational Models of Natural Argument 7 Floris Bex, Floriana Grasso, Nancy Green (eds) 16th July 2017, London, UK in [Green, 2016] were implemented in terms of To use the implemented rules for argument domain-specific semantic predicates that could in mining, i.e., to extract individual arguments, it is theory be automatically extracted by BioNLP tools. assumed that, firstly, BioNLP tools would be applied (Results of a preliminary study of human analysts’ to a source text to create a knowledge base (KB). ability to apply the argumentation schemes Named entity recognition tools such as ABNER consistently will be reported in the future.) We expect [Settles, 2005] or MutationFinder [Caporaso et al., that these rules, while domain-specific, are applicable 2007] could be used to recognize expressions to the large body of research articles on genetic referring to semantic class names such as genes, variants with effects on human health. mutations, proteins, and phenotypes. Domain- To illustrate a few of the schemes, first, the specific relations in the argumentation schemes such Method of Agreement scheme can be paraphrased as as have_phenotype and have_genotype could be follows. extracted from the text using relation extraction tools such as OpenMutationMinder [Naderi and Witte, Premises: 2012] and DiMeX [Mahmood et al., 2016]. Also, a • A group of individuals G have atypical certain amount of domain knowledge would be phenotype P required, e.g., for the relations similar and difference, • All of the individuals in G have atypical which could be acquired from a domain ontology or genotype M. domain experts. After a KB has been created, the Conclusion: M may be the cause of P (in G). argument scheme rules would be applied to the KB to recognize the premises, conclusion, and In the rules, genotype describes a variation (mutation) argumentation scheme of each argument in the text. at the level of chromosome, region on the The rules were tested by manually creating a KB and chromosome, or gene that may have a deleterious then applying the rules to the KB. effect (or effects), and phenotype describes the To illustrate, the implementation of the effect(s). This scheme can be seen as a specialization argumentation scheme for Method of Agreement is as of a more general scheme, related to Mill’s Method follows. of Agreement [Jenicek and Hitchcock, 2005]. Another scheme, related to Mill’s Method of arg( Difference [Jenicek and Hitchcock, 2005], can be scheme('Agreement'), paraphrased as follows. premise(have_phenotype(G, P)), premise(have_genotype(G, M)), Premises: conclusion(cause(M, P))) • A group of individuals G have atypical :- phenotype P group(G), • All of the individuals in G have atypical have_phenotype(G, P), genotype M. have_genotype(G, M). • A group of individuals Control do not have P. Applying this rule to a KB containing the facts: • None of the individuals in Control have M. Conclusion: M may be the cause of P (in G). group(mice1). have_phenotype(mice1, ataxia). The following scheme can be seen as a specialization have_genotype(mice1, ‘Itpr1 opt/opt’). of Argument by Analogy, e.g. as described in [Walton et al., 2008]. would derive an argument whose scheme is identified as Agreement, whose premises are the above listed Premises: facts, and whose conclusion is that the Itpr1 opt/opt • Phenotype P1 of group G1 is similar to variant may be the cause of their ataxia. Note that phenotype P2 of group G2 the rules are formulated in such a way that even • Genotype M1 of group G1 may be the cause implicit conclusions of arguments can be recognized, of P1. given the premises and an argumentation scheme • Genotype M2 of group G2 is similar to rule. The inferred conclusion itself can be added to genotype M1. the KB, enabling it to be used as a premise in Conclusion: M2 may be the cause of P2 (in G2). subsequent arguments. Also note that the conclusions of the schemes are not asserted with complete certainty. The corresponding arguments in the source 8 18th Workshop on Computational Models of Natural Argument Floris Bex, Floriana Grasso, Nancy Green (eds) 16th July 2017, London, UK text range in force from ‘plausible hypothesis’ to of an experiment, e.g. Conclusion, will be ‘fairly certain conclusion’. For details of the Prolog capitalized, and labels of argument components will implementation of the seven schemes, see [Green, be italicized.) 2016]. In the report of Experiment 1, the conclusion of This approach is in contrast to machine-learning Argument 1 must be inferred to understand the approaches to argumentation scheme recognition that Conclusion of the experiment. Also, it can be seen use only superficial text features such as keywords, that the conclusion of Argument 1 is a premise parts of speech, and clause length [Feng and Hirst, needed in the argument (Argument 2) for the 2011; Lawrence and Reed, 2016]. In addition, some Hypothesis of Experiment 2 in the next paragraph. of those approaches, e.g. [Feng and Hirst], assume Note that Experiment 2 contains two arguments, one that clauses of a text are labeled as premise or (Argument 2) for the Hypothesis, and one (Argument conclusion before argumentation scheme recognition 3) for the Conclusion. The conclusions of Argument begins. On the other hand, similar to our approach, 2 and Argument 3 are not identical, i.e., the [Saint-Dizier, 2012] uses manually-derived rules conclusion of Argument 3 is more specific than that encoded in a logic programming language for of Argument 2. The conclusion of Argument 3 can automatic identification of arguments giving reasons be challenged by a critical question of the Method of for a conclusion in instructional texts or opinion Agreement scheme, i.e., whether the putative cause texts. However, the rules are based on syntactic of the disorder is causally plausible. Experiment 3 patterns and lexical features. contains Argument 4, whose conclusion addresses that critical question. Note that, in general, critical 3 The Results Narrative questions associated with argumentation schemes provide ways in which arguments may be challenged Having proposed an approach to mining individual [Walton et al., 2008]. arguments, the next step of our research is to Paragraph 4 states a new Goal: to discover any investigate how the arguments are related to other related genetic variants causing a similar disorder in aspects of discourse structure, and how the arguments humans. The article then goes on to describe another are related to each other. Previous computation- sequence of experiments towards that goal. oriented investigations of discourse in the natural In Experiment 4, Background Knowledge and the sciences have addressed automatic classification of conclusion of Argument 4 (in Experiment 3) are used text segments, e.g., discourse coherence relations in (in Argument 5) to argue for the Hypothesis of the corpora such as BioDRB [Prasad et al., 2011] and experiment. Based on the Result of Experiment 4, BioCause [Mihaila et al., 2013], argumentative zones the implicit Conclusion (conclusion of Argument 6) [Teufel, 2010], and activities in a scientific that a deletion in ITPR1-SUMF1 may be the cause, is investigation (CoreSC) [Liakata, 2012]. None of broader than the original Hypothesis, that a deletion those annotation schemes treat arguments in the sense in ITPR1 may be the cause. However, the conclusion described in the previous section. is also narrower in the sense that it is restricted to The Results section of the article whose arguments particular individuals in the AUS1 family. were analyzed in [Green, 2016] reports on a logical In Experiments 5 and 6, the respective conclusions and temporal sequence of experiments. Arguments of Argument 7 and Argument 8 agree with the are given in the context of this narrative, i.e. the conclusion of Argument 6. Experiment 7 provides report of the scientific investigation. Figure 1 shows support (Argument 9) for a related conclusion. The our ad hoc analysis of the narrative using descriptive conclusions of arguments in Experiments 5-7 are terms similar to those of the argumentative zone and used as premises to argue for a more general CoreSC systems. The relevant content of the article conclusion in Argument 10. Then, using results of has been paraphrased in the figure. Previous Research, the authors argue (Argument 11) The Results section begins, in its first paragraph, against part of the conclusion of Argument 10. The with a description of the fortuitous discovery of an conclusion of Argument 11 is a premise of Argument inherited disorder in mice bred in the authors’ lab. 12, whose conclusion is a refinement of the Then the authors describe a sequence of three conclusion of Argument 10. Finally, the conclusion experiments intended to reveal the genetic variant of Argument 13 addresses a critical question of responsible for that mouse disorder. Figure 1 Argument 12, a relationship like that of Argument 4 describes each individual experiment in terms of its to Argument 3. Goal, (use of) Previous Research or Background As for the first of the above research questions Knowledge, Hypothesis, Method, Result, and/or (how arguments are related to other aspects of Conclusion. (To avoid confusion with argument discourse structure), the above analysis raises some components, the first letter of terms describing parts interesting possibilities. It could be that some 18th Workshop on Computational Models of Natural Argument 9 Floris Bex, Floriana Grasso, Nancy Green (eds) 16th July 2017, London, UK argumentation schemes are more commonly used for 2001] has a key feature in common with the scientific certain scientific purposes and not others. For article that we analyzed. In a discovery dialogue, the example, Argument by Analogy is used only in goal is not to try to prove or disprove a given claim, arguments for Hypotheses in the article that was but to discover something not previously known. analyzed. This could be verified by statistical analyses of corpora, and if true, could augment the 4 Discussion semantic method we have proposed for extracting individual arguments. Another possibility is that the Previous argumentation mining research has not location in the narrative could be used as a constraint addressed the natural sciences. However, on argument scheme recognition. Although it can be argumentation is an important feature of scientific seen in Figure 1 that argument content is not in one- discourse. This paper proposes a semantic approach to-one correspondence with paragraph or Experiment to automatic recognition of premises, conclusion, and boundaries, in cases where multiple rules match, a argumentation scheme of arguments in scientific text. heuristic strategy of preferring local content might be Argumentation schemes are implemented as logic applied. programs. The logic programs would be used with a As for the second of the above research questions knowledge base that could be constructed from a text (how the arguments are related to each other), the in a large part automatically using existing language analysis in Figure 1 shows that a richer model is processing tools (as described in section 2). The logic needed than support-attack relationships represented programs can be used, not only to recognize fully in current argument mining approaches, e.g., [Cabrio explicit arguments in the text, but also arguments and Villata, 2012], [Peldszus and Stede 2016], [Stab with implicit conclusions. This is important because and Gurevych, 2014]. The relationships between often the conclusions are implicit and may function conclusions of the arguments in Figure 1 is as implicit premises of subsequent arguments in the summarized in Table 1. text. Although the argumentation schemes have been implemented using domain-specific predicates, they Table 1. Dialectical structure of Results section. are specializations of more general schemes Conclusion of Relationship to Conclusion of applicable to other qualitative causal domains in the argument # argument # natural sciences. 2 Refines 1 Also, as a step towards automatic recognition of 3 Refines 2 the structure of argumentation in scientific discourse, 4 Responds to CQ 3 we present a discourse analysis of part of a scientific of text and discuss the relationship of the individual 5 Analogous to 3 arguments to other aspects of the discourse structure, 6 Broadens from 5 and how the arguments are related to each other, i.e., del. of ITPR1 to the argumentation structure of the discourse. It is ITPR1-SUMF1 shown that a richer model is needed than support- & restricts from attack relationships. humans to AUS1 Acknowledgments 7 Agrees with 6 The analysis of argument schemes was done with the 8 Agrees with 6 help of graduate students, Michael Branon and 9 Extends to more 6 Bishwa Giri, who were supported by a University of individuals North Carolina Greenboro 2016 Summer Faculty 10 Extends 9 and 6 Excellence Research Grant. 11 Attacks part of 6 and extends to References humans [Bada et al., 2012] M. Bada, M. Eckert, D. Evans, et 12 Refines 11 al. Concept Annotation in the CRAFT corpus. BMC 13 Responds to CQ 12 Bioinformatics, 13:161, 2012. of [Bratko, 2001] Ivan Bratko. Prolog Programming It would be misleading to reduce these relationships for Artificial Intelligence. 3rd edition. Addison- to pro or con some claim. As described in research Wesley, Harlow, England, 2001. on formal dialogue games for use by software agents, the discovery dialogue [McBurney and Parsons, 10 18th Workshop on Computational Models of Natural Argument Floris Bex, Floriana Grasso, Nancy Green (eds) 16th July 2017, London, UK [Cabrio and Villata, 2012] Cabrio, E. and Villata, S. Hitchcock. Logic and Critical Thinking in Medicine. 2012. Generating Abstract Arguments: A Natural American Medical Association Press, 2005. Language Approach. In Verheij, B., Szeider, S., and Woltran, S. (eds.) Computational Models of [Lawrence and Reed, 2016] Lawrence, J. and Reed, Argument: Proceedings of COMMA 2012. C. Argument Mining Using Argumentation Scheme Amsterdam, IOS Press, 454-61. Structures. In Baroni, P. et al. (eds.) Computational Models of Argument: Proceedings of COMMA 2016. [Caporaso et al. 2007] J. G. Caporaso, W.A. Amsterdam, IOS Press, 379-90. Baumgartner Jr., D. A. Randolph, K. B. Cohen, and L. Hunter. MutationFinder: A high-performance [van de Leemput et al., 2007] J. van de Leemput, J. system for extracting point mutation mentions from Chandran, M. Knight, et al. Deletion at ITPR1 text. Bioinformatics, 23:1862-1865. Underlies Ataxia in Mice and Spinocerebellar Ataxia 15 in Humans. PLoS Genetics, 3(6) e108:1076-1082, [Cardie et al., 2015] C. Cardie et al. (Eds.) Second 2007. Workshop on Argumentation Mining. North American Conference of the Association for [Liakata, M. et al., 2012] M. Liakata. Automatic Computational Linguistics, Denver, 2015. recognition of conceptualization zones in scientific articles and two life science applications. [CRAFT] http://bionlp-corpora.sourceforge.net/ Bioinformatics 28(7), 2012. CRAFT/ index.shtml. [Mahmood et. al., 2016] Mahmood A.S., Wu, T.J., [Feng and Hirst, 2011] V.W. Feng and G. Hirst. Mazumder, R. and Vijay-Shanker, K. (2016) Classifying Arguments by Scheme. In Proceedings of DiMeX: A Text Mining System for Mutation-Disease the 49th Annual Meeting of the Association for Association Extraction. PLoS One. Computational Linguistics, pages 987-996, Portland, OR, 2011. [McBurney and Parsons, 2001] P. McBurney and S. Parsons. Chance Discovery Using Dialectical [Green, 2010] N. Green. Representation of Argumentation. In New Frontiers in Artificial Argumentation in Text with Rhetorical Structure Intelligence, T. Terano et al. (Eds.). Lecture Notes in Theory. Argumentation 24(2): 181-196. Artificial Intelligence, v.2253, Berlin, Springer Verlag, 414-424. [Green, 2015a] N. Green. Identifying Argumentation Schemes in Genetics Research Articles. In [Naderi and Witte, 2012] N. Naderi and R. Witte. Proceedings of the Second Workshop on Automated extraction and semantic analysis of Argumentation Mining, North American Conference mutation impacts from the biomedical literature. of the Association for Computational Linguistics BMC Genomics, 13(Suppl 4):510, 2012. (NAACL), Denver, CO, 2015. [Peldszus and Stede, 2016] A. Peldszus, M. Stede. [Green, 2015b] N. Green. Annotating Evidence- Rhetorical structure and argumentation structure in Based Argumentation in Biomedical Text. In Proc. monologue text. In: Proceedings of the 3rd Workshop 2015 Int. Workshop on Biomedical and Health on Argumentation Mining, ACL Berlin, 2016. Informatics, IEEE Int. Conf. on Bioinformatics and Biomedicine (BIBM 2015), Washington, D.C, Nov. 9- [Prasad et al., 2011] R.Prasad, S. McRoy, N. Frid, A. 12, 2015. IEEE Computer Society Press. Joshi, and H. Yu. The Biomedical Discourse Relation Bank. BMC Bioinformatics, 12:188, 2011. [Green, 2016] N. Green. Implementing Argumentation Schemes as Logic Programs. [Reed et al., 2016] C.Reed et al. (Eds.) Third Workshop on Computational Models of Natural Workshop on Argumentation Mining. Association for Argument (CMNA 2016). International Joint Computational Linguistics, Berlin, 2016. Conference on Artificial Intelligence, New York. [Saint-Dizier, P, 2012] P. Saint-Dizier. Process [Green et al., 2014] N. Green, et al. (Eds.) First natural language arguments with the Workshop on Argumentation Mining. Association for platform. Argument and Computation 3(1), March Computational Linguistics, Baltimore, MD., 2014. 2012, 49-82. [Jenicek and Hitchcock, 2005] M. Jenicek and D. 18th Workshop on Computational Models of Natural Argument 11 Floris Bex, Floriana Grasso, Nancy Green (eds) 16th July 2017, London, UK [Settles, 2005] B. Settles. ABNER: an open source [Walton et al., 2008] D. Walton, C. Reed, and F. tool for automatically tagging genes, proteins, and Macagno. Argumentation Schemes. Cambridge other entity names in text. Bioinformatics University Press, 2008. 21(14):3191-3192, 2005. [Stab and Gurevych, 2014] C. Stab and I. Gurevych, Annotating Argument Components and Relations in Persuasive Essays. In Proc. COLING 2014, pp. 1501- 1510. [Teufel, 2010] S. Teufel, S. The Structure of Scientific Articles: Applications to Citation Indexing and Summarization. Stanford, CA, CSLI Publications. [Verspoor et al., 2012] K. Verspoor, K.B. Cohen, A. Lanfranchi, et al. A Corpus of Full-text Journal Articles is a Robust Evaluation Tool for Revealing Differences in Performance of Biomedical Natural Language Processing Tools. BMC Bioinformatics 13:207, 2012. Appendix. Figure 1. Analysis of discourse structure and argumentation. ¶1 Report Observation: Some mice bred in the authors’ lab are affected with an autosomal recessive disorder resembling a kind of ataxia. (premise, Argument 1) ¶2 Report Experiment 1: Method: Linkage analysis Result: The affected mice have a lesion on chromosome 6qE1 (premise, Argument 1) Conclusion (implicit): A genetic variant on 6qE1 may be the cause of their disorder. (conclusion, Argument 1 - Method of Agreement; premise, Argument 2) ¶3 Report Experiment 2: Previous Research: A certain deletion on 6qE1 in gene Itpr1, the Itpr1 opt/opt variant, is known to cause a similar disorder in mice. (premise, Argument 2) Hypothesis: A deletion in Itpr1 may be the cause the disorder in the lab’s mice. (conclusion, Argument 2 – Analogy) Method: Sequence Itpr1 Result: The lab’s affected mice have a deletion in Itpr1: the Itpr1 Δ18/Δ18 variant. (premise, Argument 3) Conclusion: The Itpr1 Δ18/Δ18 variant may be the cause of their disorder. (conclusion, Argument 3 – Method of Agreement) Report Experiment 3: Previous Research: Cerebellar Purkinjee cells of Itpr1 opt/opt mice, who have ataxia, have decreased Itpr1 expression (premise, Argument 4) Method: Measure expression of Itpr1 in cerebellar Purkinjee cells of mice with the Itpr1 Δ18/Δ18 variant. Result: Decreased level of Itrpr1 expression found. (premise, Argument 4) Conclusion (implicit): There is a plausible explanation at the molecular level of how deletions in Itpr1 may cause ataxia-like disorders in mice. (conclusion, Argument 4 – Consistent Explanation; premise, Argument 5) ¶4 Report Goal: Discover cause of cognate human disorders, such as spinocerebellar ataxia 15 (SCA15), where no causal mutation has been identified. 12 18th Workshop on Computational Models of Natural Argument Floris Bex, Floriana Grasso, Nancy Green (eds) 16th July 2017, London, UK ¶5 Report Experiment 4: Background knowledge (implicit): The ITPR1 gene in humans is functionally similar to Itpr1 in mice. (premise, Argument 5) Hypothesis: A deletion in ITPR1 is a cause of SCA15 in humans. (conclusion, Argument 5 – Analogy) Method: Sequence DNA from three AUS1 family members with SCA15. Result: The three family members had a deletion in ITPR1-SUMF1. (premise, Argument 6; premise, Argument 7) Conclusion (implicit): A deletion in ITPR1-SUMF1 may be the cause of ataxia in the three AUS1 family members. (conclusion, Argument 6 – Method of Agreement) Report Experiment 5: Goal: Determine if the ITPR1-SUMF1 deletion is a benign polymorphism. Method: Compare to ITPR1 and SUMF1 in two control groups. Results: No deletion found in ITPR1 or SUMF1 in the control groups. (premise, Argument 7) Conclusion: A deletion in ITPR1-SUMF1 may be the cause of ataxia in the three affected AUS1 family members. (conclusion, Argument 7 – Method of Difference) ¶6 Report Experiment 6: Method: Fine-map the breakpoints of the deletion in the affected AUS1 family members and in the controls. Result: Deletion of the first three of the nine exons of SUMF1 and the first 10 of the 58 exons of ITPR1 in the affected family members only. (premise, Argument 8) Conclusion: A deletion in ITPR1-SUMF1 may be the cause of ataxia in the three AUS1 family members. (conclusion, Argument 8 – Method of Difference) ¶7 Report Experiment 7: Method: Analyzed two additional families (H33 and H27) with an inherited cerebellar ataxia similar to that described in the AUS1 family. Result: The affected H33 and H27 family members have a deletion at the SCA15 locus from SUMF1 through IPTR1; the unaffected family members do not. (premise, Argument 9) Conclusion (implicit): A deletion in ITPR1-SUMF1 may be the cause of the ataxia disorder in the affected H33 and H27 family members. (conclusion, Argument 9 – Method of Difference) Report Conclusion: In three families [AUS1, H33, H27] cerebellar ataxia segregated with a deletion in SUMF1-ITPR1, not observed in controls. (premise, Argument 10) The deletion in ITPR1-SUMF1 is the cause of SCA15 in those families. (conclusion, Argument 10 – Method of Difference; premise, Argument 12) ¶8 Report Previous Research: Homozygous mutation of SUMF1 results in autosomal recessive multiple sulfatase deficiency… No co-occurrence of ataxia has been described in heterozygous parents [i.e. who have one copy of a SUMF1 mutation] of patients with multiple sulfatase deficiency (premise Argument 11). It is improbable that the deletion of SUMF1 … itself causes or contributes to SCA15 (conclusion Argument 11 – Failed Method of Agreement; premise Argument 12). [Therefore, deletion in ITPR1 is the likely cause of ataxia in the three families (implicit conclusion, Argument 12 – Eliminate Difference)] Mutation of ITPR1 is biologically plausible as a cause of ataxia (conclusion, Argument 13 – Consistent Explanation 18th Workshop on Computational Models of Natural Argument 13 Floris Bex, Floriana Grasso, Nancy Green (eds) 16th July 2017, London, UK