Month: March 2018

This post is a draft of part of a piece that I’m writing at the moment, and on which I would like your feedback.  The topic is variability in language.  I pay the rent by researching the issues involved in getting computers to understand biomedical language–for example, the language of scientific journal articles, or the language of health records.  I’m in the midst of writing a chapter about this topic for a handbook of computational linguistics.  The audience is people who are interested in computational linguistics, but don’t have any experience with the biomedical domain.  If you’re a reader of this blog, that’s probably not a bad description of you.  So, it would be super-helpful to me to have your critique of this material.  I’m looking for anything that isn’t clear, anything that makes it difficult to understand my prose–anything that you think could be improved.  My grandmother will tell me how wonderful it is, so just feel free to plow into me with both fists–seriously, you’d be surprised at how much pain you can take in your old age, and I’m getting pretty old.  

Variability is the property of being able to express the same proposition in multiple ways.  If ambiguity is the major problem of natural language processing, variability is the second.  From a theoretical perspective, the field of sociolinguistics sees the study of variation in language as the central problem of linguistics, and it makes a strong case for that claim (e.g. Labov 2004)[1].  From a practical perspective in natural language processing, the high degree of variability in natural language prevents us from ever being able to use a dictionary-like data structure (such as hash tables, B-trees, or tries) to accomplish our tasks: we will never have a “dictionary” of all possible sentences (Chomsky 1959)[2].  This kind of approach would be fast and efficient—if only it were possible (Gusfield 1997)[3].

Sources of variability

 Some of the sources of variability in language are well-known even to the casual reader—for example, synonymy, or the availability of multiple words that have the same dictionary meaning.  A kind of synonymy that is especially relevant in biomedical languages occurs when there is both a technical and a lay or common term for something, such as the lay term heart attack and the technical term myocardial infarction.  Using technical terminology is important for the precision of scientific writing and of medical records (Rey 1979)[4].  However, the use of technical terminology also can make it difficult for patients and their families to learn about their illness or to understand their own health records (Kandula et al. 2010)[5].  One way to deal with this problem is to use natural language processing techniques to replace technical terms with their lay synonyms (Elhadad 2006[6], Elhadad and Sutaria 2007,[7] Deléger and Zweigenbaum 2009[8], Leroy et al. 2013a[9], Leroy et al. 2013b[10]) or their definitions (Elhadad 2006)[11] in order to make clinical documents or scientific journal articles accessible to non-professionals.  Doing this computationally, rather than manually, allows it to be done at enormous scales, or on demand.  This is a good example of why to do natural language processing in the biomedical domain: the possibility of doing real good in the world.

Paraphrase is the phenomenon of different (and typically syntactically different) expressions in language of the same meaning (Ganitkevitch et al. 2013)[12].   Where synonymy operates of the level of words, paraphrase operates at the level of the phrase, or group of words.  Paraphrasing is a source of variability that is especially interesting in the biomedical domain because of how it interacts with the technical vocabulary of the field (Deléger and Zweigenbaum 2008, Deléger 2009, Deléger and Zweigenbaum 2010, Grabar and Hamon 2014)[13],[14],[15],[16]. Funk et al. looked for possibilities to paraphrase or replace synonyms in 41,853 terms from the Gene Ontology, and found that 27,610 out of 41,852 were paraphrasable, or had synonyms, or both[17].  This indicates that the possibilities for variant forms of the same thing occurring in the biomedical literature are tremendous.

But, do those tremendous numbers of variants really occur?  It appears that they do.  Cohen et al. (2008) looked at the incidence of alternative syntactic constructions involving common nominalizations (nouns derived from verbs, such as treatment from to treat) in scientific journal articles—for example, drug treatment of cancer and cancer treatment with drugs.  Figure 1 shows a typical finding: for some nominalizations, as many as 15 out of 16 possible variants could be found even in a relatively small corpus[18].

How different can these paraphrases be from each other?  Technical terms in biomedical research can be quite long, which means that there can be multiple candidates for paraphrasing and for replacement of synonyms (see above).  This means that the number of possible paraphrases of a long term can be explosive.  Those paraphrases, even for a short term, can be quite different—for example, Cohen et al. (2017) examined the relationship between the length of terms in the Gene Ontology and the length of appearances of those terms in the CRAFT corpus of biomedical journal articles, and found that 2-word terms could show up with paraphrases as long as 15 words[19].  The high incidence of just these two forms of variability in language—synonymy and paraphrasing—as well as the large differences that can be seen in forms with the same meanings illustrate just how much of an issue variability is for natural language processing in general, and in biomedical texts in particular.

Harsh critiques in the Comments section below, please!

[1] Labov, William. “Quantitative reasoning in linguistics.” Sociolinguistics/Soziolinguistik: An international handbook of the science of language and society 1 (2004): 6-22.

[2] Chomsky, Noam. “A review of BF Skinner’s Verbal Behavior.” Language 35, no. 1 (1959): 26-58.

[3] Gusfield, Dan. Algorithms on strings, trees and sequences: computer science and computational biology. Cambridge university press, 1997.

[4] Rey, Alain. La terminologie: noms et notions. No. 1780. Presses Univ. de France, 1979, p. 56.

[5] Kandula, Sasikiran, Dorothy Curtis, and Qing Zeng-Treitler. “A semantic and syntactic text simplification tool for health content.” In AMIA annual symposium proceedings, vol. 2010, p. 366. American Medical Informatics Association, 2010.

[6] Elhadad, Noemie. “Comprehending technical texts: Predicting and defining unfamiliar terms.” In AMIA annual symposium proceedings, vol. 2006, p. 239. American Medical Informatics Association, 2006.

[7] Elhadad, Noemie, and Komal Sutaria. “Mining a lexicon of technical terms and lay equivalents.” In Proceedings of the Workshop on BioNLP 2007: Biological, Translational, and Clinical Language Processing, pp. 49-56. Association for Computational Linguistics, 2007.

[8] Deléger, Louise, and Pierre Zweigenbaum. “Extracting lay paraphrases of specialized expressions from monolingual comparable medical corpora.” In Proceedings of the 2nd Workshop on Building and Using Comparable Corpora: from Parallel to Non-parallel Corpora, pp. 2-10. Association for Computational Linguistics, 2009.

[9] Leroy, Gondy, David Kauchak, and Obay Mouradi. “A user-study measuring the effects of lexical simplification and coherence enhancement on perceived and actual text difficulty.” International journal of medical informatics 82, no. 8 (2013): 717-730.

[10] Leroy, Gondy, James E. Endicott, David Kauchak, Obay Mouradi, and Melissa Just. “User evaluation of the effects of a text simplification algorithm using term familiarity on perception, understanding, learning, and information retention.” Journal of medical Internet research 15, no. 7 (2013).

[11] Elhadad, Noemie. “Comprehending technical texts: Predicting and defining unfamiliar terms.” In AMIA annual symposium proceedings, vol. 2006, p. 239. American Medical Informatics Association, 2006.

[12] Ganitkevitch, Juri, Benjamin Van Durme, and Chris Callison-Burch. “PPDB: The paraphrase database.” Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2013.

[13] Deléger, Louise, and Pierre Zweigenbaum. “Paraphrase acquisition from comparable medical corpora of specialized and lay texts.” AMIA Annual Symposium Proceedings. Vol. 2008. American Medical Informatics Association, 2008.

[14] Deléger, Louise. Exploitation de corpus parallèles et comparables pour la détection de correspondances lexicales: application au domaine médical. Diss. Paris 6, 2009.

[15] Deléger, Louise, and Pierre Zweigenbaum. “Identifying Paraphrases between Technical and Lay Corpora.” LREC. 2010.

[16] Grabar, Natalia, and Thierry Hamon. “Unsupervised method for the acquisition of general language paraphrases for medical compounds.” Proceedings of the 4th International Workshop on Computational Terminology (Computerm). 2014.

[17] Funk, Christopher S., K. Bretonnel Cohen, Lawrence E. Hunter, and Karin M. Verspoor. “Gene Ontology synonym generation rules lead to increased performance in biomedical concept recognition.” Journal of biomedical semantics 7, no. 1 (2016): 52.

[18] Cohen, K. Bretonnel, Martha Palmer, and Lawrence Hunter. “Nominalization and alternations in biomedical language.” PloS one 3.9 (2008): e3158.

[19] Cohen, K. B., Verspoor, K., Fort, K., Funk, C., Bada, M., Palmer, M., & Hunter, L. E. (2017). The colorado richly annotated full text (craft) corpus: Multi-model annotation in the biomedical domain. In Handbook of Linguistic Annotation (pp. 1379-1394). Springer, Dordrecht.

Harsh critiques in the Comments section below, please!

I pay the rent by researching the issues involved in getting computers to understand biomedical language–for example, the language of scientific journal articles, or the language of health records.  I’m in the midst of writing a chapter about this topic for a handbook of computational linguistics.  The audience is people who are interested in computational linguistics, but don’t have any experience with the biomedical domain.  If you’re a reader of this blog, that’s probably not a bad description of you.  So, it would be super-helpful to me to have your critique of my introduction.  I’m looking for anything that isn’t clear, anything that makes it difficult to understand my prose–anything that you think could be improved.  My grandmother will tell me how wonderful it is, so just feel free to plow into me with both fists–seriously, you’d be surprised at how much pain you can take in your old age.  

What makes the biomedical domain an interesting one from the perspective of computational linguistics?  Indeed, what makes any domain an interesting one from the perspective of computational linguistics?  In fact, Roger Shuy has asserted that the notion of any specific kind of data defining a particular area of linguistics is unsupportable.  As he puts it: “There is little reason for the data on which a linguist works to have the right to name that work” (Shuy 2002)[1].

Shuy’s statement is surprising, since he himself is North America’s leading forensic linguist—a linguist whose career has been defined entirely by his excellent work on language as it appears in the legal system.  And, indeed, many computational linguists describe themselves as doing biomedical natural language processing[2].

So, why study computational linguistics in the biomedical domain?  One can identify at least three primary types of reasons: theoretical, practical, and use-case-oriented.

Theoretical aspects of biomedical language

Biomedical languages are of interest to computational linguistics for two reasons: their relevance to questions about the nature and limits of grammar, and the light that they can shed on issues of reproducibility in natural language processing.

Biomedical languages and grammaticality

Biomedical languages are of interest from the perspective of computational linguistics in part because they stretch the limits of what can possibly be grammatical in a natural language.  Since the second half of the 20^th century, much of linguistic argumentation has focused around grammaticality, which at a first approximation we can define as the question of whether or not an utterance is within the boundaries of some language, or not (Partee et al. 2012).  Early in the second half of the 20^th century, utterances that came under discussion in linguistic debates tended to be either quite ordinary (such as the famous John loves Mary (Fowle 1850)[3]), or interestingly ambiguous—sentences like John loves his wife, and so does Tom (Duží 2012[4]) whose grammaticality (as opposed to their interpretations) was mostly not in question.  Although the discourse of that period of linguistic inquiry—particularly with respect to the development of syntactic theory—was often couched in terms of defining—and constraining—some set of sentences (“strings”), in practice it tended to be more about operations on (and to a much lesser extent, interpretation of) those strings.

This changed in the 1970s and1980s with the emergence of a research community that explored sublanguages: language associated with a particular genre and a particular kind of interlocutor[5].  Harris (1976) laid out a number of the principles of the sublanguage approach: semantics was embraced, not pushed off to some later date[6].  Although not always formalized as such, lexical preferences and statistical tendencies were taken advantage of (unusual in the era of a linguistics that had a complicated relationship with the lexicon and famously open disdain for statistics (Harris 1995)[7]). As Grishman (2001) explains, these were interesting for at least two reasons: they seemed amenable to syntactic description by reducing complex syntactic structures into simpler ones, reminiscent to the transformational analyses that were becoming dominant in linguistics, and they held the promise of mapping to a tractable model of the world, or semantics[8]—something that had largely eluded linguistics up to that point[9].

The biomedical domain seemed like a fruitful area of research to the early investigators of the topic, and it was.  Scientific journal articles were one such genre, with the interlocutors being researchers; clinical documents provided another, with the interlocutors being physicians.  Harris et al. (1989) provided an in-depth description of the language of scientific publications about immunology [10].  It set a standard for sublanguage research on biomedical languages that would remain unparalleled for years.  The usefulness of the sublanguage model can be seen in the fact that researchers continue to find it fruitful (some prominent examples in the biomedical domain are reviewed in Demner-Fushman et al. 2009)[11].  Some examples that illustrate particularly well the use of the sublanguage model for semantic representation include Dolbey (2009) in the molecular biology domain[12] and Deléger et al. (2017), which also includes a review of the basic issues and of other approaches to resolving them[13].  Clinical sublanguages soon turned out to be full of data that was ungrammatical on any standard treatment of syntax (see Table 1 for some examples), making it clear that they were good areas for investigating the limits of grammaticality at a time when grammaticality was generally considered a binary characteristic of language with strict semantic constraints .

Table 1: Examples of ungrammatical sentences from radiology reports.  In English, the verb to show is usually thought of as requiring a sentient subject.  In these sentences, we see a wide range of non-sentient subjects: an anatomical organ (chest), an event (examination), x-ray films (x-rays of spine), a laboratory test (urinalysis), and the output of a computed tomography exam (brain scan).  All of the sentences have “generic” noun phrases where they would normally require an article or demonstrative (chest, examination, x-rays of spine, and brain scan).  Source: Hirschman (1986)[14].  No human subjects approval or HIPAA training is required for use of these examples.

[1] Shuy, Roger. Linguistic battles in trademark disputes. Springer, 2002.

[2] The Association for Computational Linguistics Special Interest Group on Biomedical Natural Language Processing has over 100 members at the time of writing.

[3] Fowle, William B. (1850) “English Grammar: Goold Brown.” Common School Journal, pp. 245-249.

[4] Duží, Marie (2012) “Extensional logic of hyperintentions.”  In Düsterhöft, Antje, Meike Klettke, and Klaus-Dieter Schewe, eds. Conceptual Modelling and Its Theoretical Foundations: Essays Dedicated to Bernhard Thalheim on the Occasion of His 60th Birthday. Vol. 7260. Springer Science & Business Media, 2012.

[5] See Chapter 18, Sublanguages and controlled languages, this volume.

[6] Harris, Zellig. “On a theory of language.” The Journal of Philosophy 73.10 (1976): 253-276.

[7] Harris, Randy Allen. The linguistics wars. Oxford University Press, 1995.

[8] Grishman, Ralph. “Adaptive information extraction and sublanguage analysis.” Proc. of IJCAI 2001. 2001.

[9] Harris, Randy Allen. The linguistics wars. Oxford University Press, 1995.

[10] Harris, Z., Gottfried, M., Ryckman, T., Daladier, A., & Mattick, P. (2012). The form of information in science: analysis of an immunology sublanguage (Vol. 104). Springer Science & Business Media.

[11] Demner-Fushman, Dina, Wendy W. Chapman, and Clement J. McDonald. “What can natural language processing do for clinical decision support?.” Journal of biomedical informatics 42.5 (2009): 760-772.

[12] Dolbey, Andrew. “BioFrameNet: a FrameNet extension to the domain of molecular biology.” (2009).

[13] Deléger, Louise, Leonardo Campillos, Anne-Laure Ligozat, and Aurélie Névéol. “Design of an extensive information representation scheme for clinical narratives.” Journal of biomedical semantics 8, no. 1 (2017): 37.

[14] Hirschman, Lynette. “Discovering sublanguage structures.” Analyzing Language in Restricted Domains: Sublanguage Description and Processing (1986): 211-234.

Harsh critiques in the Comments section below, please!