1. Representation of biomedical sublanguages using symbolic notation
John MacMullen
SILS Bioinformatics Journal Club
Fall 2003

2. Terminology article assumptions
“Knowledge encoded in textual documents is organized around sets of domain-specific terms…” [938]
“Terms represent the most important concepts in a domain and characterize documents semantically.” [939]
“[T]he basic problem is to recognize domain-specific concepts and to extract instances of specific relationships among them.” [938]
Terms are ambiguous and have variation; they are hardly ever mono-referential
The lack of naming conventions (controlled vocabularies), the existence of acronyms, and the large existing heterogeneous literatures increase complexity.
[from Nenadic, G., Spasic, I., & Ananiadou, S. (2003). Terminology-driven mining of biomedical literature. Bioinformatics 19(8), ]
SILS Bioinformatics Journal Club – Fall 2003

3. SILS Bioinformatics Journal Club – Fall 2003
Harris’ Assumptions
“[T]here is a particular structure to science information in general, and to the information of each subscience in particular”, [because] “for each subscience there are particular subsets of nouns that occur with particular subsets of verbs or other words” [215].
“[I]t is not intrinsic properties of sounds and meanings that determine the possible word-sequences of sentences.” […] “For each word, we find roughly stable inequalities of probability among the words in its required (positive probability) set” [216].
“What is common to the texts of a given subject matter is that first-level words of a given subset require zero-level words of only a particular subset” [217].
“We thus obtain for the science several statement-types […]” [217].
“What we have here is thus an information-theoretic approach to the structure of information, as against solely the amount of information” [217, emphasis added]
SILS Bioinformatics Journal Club – Fall 2003

4. Linguistic probabilities
“For each word, we find roughly stable inequalities of probability among the words in its required (positive probability) set” [216].
“The meaning of a word is indicated, and in part created, by the meanings of the words in respect to which it has higher than average probability” [217]
“Words with highest probability in respect to another word, or which otherwise can be shown structurally to have highest expectancy, add little or no information” [217].
SILS Bioinformatics Journal Club – Fall 2003

5. Representing Meaning with Notation
Movement towards structured rather than natural language
Representing sentences as propositions whose truth can be tested
Example [from ]:
If ‘G’ = “antigen”,
and ‘J’ = “injected into”,
and ‘B’ = “ear”,
then ‘GJB’ = “antigen injected into ear”
SILS Bioinformatics Journal Club – Fall 2003

6. Symbolic representation
SILS Bioinformatics Journal Club – Fall 2003

7. SILS Bioinformatics Journal Club – Fall 2003
Applications
Investigate “the possibilities of obtaining standard notations for science languages,not by fiat but by boiling down from actual use…” [220].
“[R]elate the information structure of a science to anything else that characterizes the field, in order to reach if possible a ‘‘structure’’of the science” [220].
“[S]ee how tabular or other two-dimensional displays can represent the data (or the Result statements) of articles, for human inspection or for computer processing” [220].
SILS Bioinformatics Journal Club – Fall 2003

8. SILS Bioinformatics Journal Club – Fall 2003
Other Propositions
“[W]hen, in a given science, articles written in different languages are analyzed […], we obtain the same sentence-types and structures,with only small differences due to the languages.
“The word class and subclass symbols,and the sentence-types,are therefore not just a sublanguage of a particular language, but an independent symbolic linguistic system” [219].
Difference between ‘equivalance’ and ‘equal’?
Example:
La casa di Gianni è bianco. [original]
The house of Gianni is white. [literal or equal]
John’s house is white. [equivalent]
SILS Bioinformatics Journal Club – Fall 2003

9. SILS Bioinformatics Journal Club – Fall 2003
Questions
Assume Harris’ notation method is valid and works well.
How might it be implemented in practice? (This is both an algorithm question and a policy question.)
Who would apply it?
What would some of the barriers be?
Do Harris' arguments hold in an interdisciplinary environment?
SILS Bioinformatics Journal Club – Fall 2003

10. SILS Bioinformatics Journal Club – Fall 2003
References
Harris, Zellig S. (2002). The structure of science information. Journal of Biomedical Informatics, 35,
Linguistic String NYU:
MedLEE project (Medical Language Extraction and Encoding System):
Zellig Harris homepage:
SILS Bioinformatics Journal Club – Fall 2003