Semantics and Knowledge Organization by Birger Hjørland Visiting professor Riga, October 2006 Time: (2 x 45 min)

2 Contents 1.Semantics exemplified by thesauri 2.Semantics and other kinds of knowledge organizing systems/ 3.Semantic tools 4.Theories of semantics 4b. Kinds of semantic relations 5. Implications for Library and Information Science (LIS)

3 1. Semantics exemplified by thesauri What is semantics? Let us start by looking at thesauri. A thesaurus is typically: 1)A selection of concepts 2) An indication of synonyms (”USE”/”USED FOR”) 3) An indication of homonyms (parenthetical qualifiers, e.g.: Letters (Alphabet); Letters (Correspondence); 4) An indication of generic relations (”BT”, ”NT”) (these may also be part-whole relations).

4 1. Semantics exemplified by thesauri 5) An indication of ”related terms” (”RT”) which may be other kinds of semantic relations such as antagonistic relations, active/passive relations, or causal relations. 6) A ” scope note ” defining the term. “A good thesaurus provides, through its hierarchy augmented by associative relationships between concepts, a semantic road map for searchers and indexers and anybody else interested in an orderly grasp of a subject field.” (Soergel, 2004).

5 1. Semantics exemplified by thesauri We may thus conclude that thesauri are essentially a kind of semantic tools, which provides ”a semantic road map”. The semantic relations provided are relations between concepts or meanings, such as BT/NT & RT. We have now an idea of the meaning of the term ”semantics”.

6 1. Semantics exemplified by thesauri Semantic relations are the relations between concepts, meanings or senses. The concept [school] should be distinguished from the word ‘school’. [School] is a kind of [educational institution]. This indicates a hyponymous or hierarchical relationship between two concepts or meanings, which is one kind among a long range of kinds of semantic relations.

7 1. Semantics exemplified by thesauri The concept [School] may, for example, be expressed by the terms or expressions ‘school’ ‘schoolhouse’ and ‘place for teaching’. The relation between ‘school’ and ‘schoolhouse’ is a (synonym) relation between two words, while the relation between ‘school’ and ‘place for teaching’ is a relation between a word and an expression. The relations between words are termed lexical relations 'School' also means [a group of people who share common characteristics of outlook, a school of thought]. This is a homonym relation: Two senses share the same word or expression: ‘school’. Synonyms and homonyms are not relations between concepts, but are about concepts expressed with identical or with different signs.

8 1. Semantics exemplified by thesauri Relations between concepts, senses or meanings should not be confused with relations between the terms, words, expressions or signs that are used to express the concepts. It is, however, common to mix both of these kinds of relations under the heading "semantic relations". I follow this tradition and this makes it even more clear that thesauri are nothing but semantic tools.

9 2. Knowledge organizing systems (KOS) Term Lists  Authority Files  Glossaries  Dictionaries  Gazetteers Classifications and Categories  Subject Headings  Classification Schemes  Taxonomies  Categorization Schemes Relationship Lists  Thesauri  Semantic Networks Ontologies

10 2. Knowledge organizing systems (KOS) The list of KOS at the former slide was provided by Hoge (2000). All the KOS mentioned here are also semantic tools. For example, an authority list is a list of ”preferred terms”, i.e. a list providing information about preferred and non-preferred words, names or symbols. It is a semantic tool providing less information about semantic relations compared to thesauri.

11 2. Knowledge organizing systems (KOS) Classification systems are also semantic tools. Each class represent a concept. One difference is that classifications systems usually use a non-verbal notation. Another difference is that their arrangement is primarily systematic rather than alphabetically and a third difference is that they tend to divide the universe of knowledge in a top-down fashion rather than concept by concept in a bottom-up way.

12 2. Knowledge organizing systems (KOS) Nonetheless, the classes represent concepts, and the hierarchical structure of the classes represent a kind of semantic relation (mostly the generic relation). A classification also often has notes corresponding to the scope notes in thesauri. It is thus justified to regard classification systems as semantic tools. It is my claim that the same can be said about all KOS in Hodge (2000). (Se further in Lifeboat for Knowledge organization).

13 2. Knowledge organizing systems (KOS) Semantic networks and ontologies are kinds of semantic tools displaying much richer semantic relations compared to, for example, thesauri. Some modern tools, like topic maps, makes it possible to define the kinds of relations one wants to. Thesauri and classification systems may be made with topic maps technology if only few kinds of relations are utilized. One sometimes speak of the need to enriching thesauri with more kinds of semantic relations. Such considerations blur the difference between different kinds of semantic tools.

14 2. Knowledge organizing systems (KOS) I should be said, however, that I consider the concept of KOS to be broader than specified by Hodge (2000). An encyclopedia is also a kind of KOS. So is a scientific discipline and a profession and many other things. Such kinds of KOS should not be considered semantic tools, although, for example, Umberto Eco argue that a dictionary is a ” disguised encyclopedia”.

15 3. Semantic tools The generic term ”semantic tools” is important because it allows us to generalize and to base the theory of classifications, thesauri, semantic networks, ontologies etc. on a common theoretical base related to a theory of semantics. In other words: It provides a hit on where to look for a theoretical understanding of such systems. (Whether or not fruitful semantic theories a readily available is quite another matter!)

16 4. Theories of semantics The first thing to consider is that a given semantic relation between two terms is seldom constant i all contexts. This is somewhat surprising to realize because we may look semantic relations up in dictionaries and other kinds of semantic tools. We are used to consider, for example, ”computer” and ”data processor” as well as ”pirate” and ”sea robber” as synonyms. It is a strange thought that such relations are relative to context. The famous semiotic triangle (Ogden & Richard, 1923) expresses this relativity:

17

18 4. Theories of semantics Or, with the terminology of Peirce: "A sign, or representamen, is something which stands to somebody for something in some respect or capacity. It addresses somebody, that is, creates in the mind of that person an equivalent sign, or perhaps a more developed sign. That sign which it creates I call the interpretant of the first sign. The sign stands for something, its object [or referent]. It stands for that object, not in all respects, but in reference to a sort of idea, which I have sometimes called the ground of the representamen." (Peirce, , 2, 228).

19 4. Theories of semantics It is important for a theory of semantics to realize both the relative stability and the context-dependence of semantic relations. It is important to ask: When is a semantic relation? When is it fruitful to consider two terms semantically related in a given way? (and how do we determine when it is fruitful to consider two terms semantically related in a given way?)

20 4. Theories of semantics Hjørland (2007) discuss the following possibilities. Semantic relations may be: : a) Query/situation specific or idiosyncratic b) Universal, Platonic entities/relations c) “Deep semantics” common to all languages (or inherent in cognitive structures) d) Specific to specific empirical languages (e.g. Swedish) e) Domain or discourse specific f) Other (e.g. determined by a company or by a workgroup, “user oriented”)

21 4. Theories of semantics a) Query/situation specific or idiosyncratic ""When I use a word," Humpty Dumpty said, in rather a scornful tone, "it means just what I choose it to mean— neither more nor less." "The question is, "said Alice, "whether you can make words mean so many different things." "The question is," said Humpty Dumpty "which is to be master—that's all."" Carroll (1899)

22 4. Theories of semantics It is important to keep in mind that concept determination and semantic relations are to be used in, for example, query expansion (automatic or manual) as well as in query precision and query formulation. In a way it is the specific "information need" that determines which relations are fruitful and which are not fruitful in a given search session. A semantic relation that increases recall and precision in a given search is relevant in that situation. Creative information searchers do just that: They provide search strategies that bring forward a fruitful set of identified documents by combining terms in unusual ways.

23 4. Theories of semantics On the other hand is it clear that if we base a semantic theory on an individual/idiosyncratic view of concepts and semantics, then it is not possible to design systems for more than one user or situation, which is absurd. We need more stable principles on which to determine semantic relations. We need a semantic theory about the meaning of words in forms of typified practices. Knowledge about semantics in typified practices may then be used by information searchers in one way or another in order to include or exclude certain documents.

24 4. Theories of semantics b) According to Platonism, the meaningfulness of a general term is constituted by its connection with an abstract entity, the (possibly) infinite extension of which is determined independently of our classificatory practices. (cf., Haukioja, 2005). It may be sufficient to say that the semantics of, for example, mathematical concepts are not just given by intuition by the individual indexer. They have to be determined by considering the mathematical literature (or people educated in that literature). Even if the basic method in mathematics is a kind of rational intuition this does not imply that semantic relations in mathematic should be considered to be given a priory in knowledge organization.

25 4. Theories of semantics c) A certain version of “deep semantics” is the theory of semantic primitives according to which every word can be broken up into primitive kernels of meaning, semantemes (also called semantic features or semantic components). Such theories have been influential in thesaurus construction and also in theories of dictionaries. They may still be influential under labels such as ”formal concept analysis”. Also facet-analysis are related to this semantic view.

26 4. Theories of semantics WordSemantemes Fathermale + parent Motherfemale + parent Sonmale + offspring Daughterfemale + offspring Brothermale + sibling Sisterfemale + sibling

27 4. Theories of semantics d) Often are semantic associated with specific languages. A paper by Hedlund et al. (2001) is titled “Aspects of Swedish morphology and semantics from the perspective of mono- and cross-language information retrieval”. The implication of what is said in the title is that the Swedish language has a semantics of its own. In other words: Semantic relations are structural relations attributed to different empirical languages.

28 English*German*Danish*French Italian Spanish TreeBaum Træ arbre albero Árbol WoodHolz bois legno Leña skov Madera Woods Wald bosco Bosque Forestforêt foresta Selva

29 4. Theories of semantics e) The kind of semantic theory that I believe is fruitful (and true) is based on metatheoretical principles such as: The focus on the relation between meaning and the real world (realism) The focus on the functional/pragmatic nature of meaning (pragmatism) The focus on the development of meaning in a social context (historicism and meaning collectivism/holism). It is based on pragmatic philosophers such as Charles Sanders Peirce, John Dewey, Hilary Putnam and on the Russian ”activity theory”.

30 4. Theories of semantics The pragmatic view of concepts may be understood as follows: To be in order to pursue goals and to plan future actions must intelligent systems – humans or machines – be able to classify some objects, behaviors or events as equivalent for achieving given goals. Concepts are ways of classifying the world. Any concept/classification may be more or less suited or unsuited helping the system achieving a particular goal.

31 4. Theories of semantics If a human culture, for example, do not conceptualize animals as eatable, that culture have fewer objective possibilities to feet its members compared to other cultures. An intelligent system must be able to form classes of objects that are equivalent in relation to a certain task. A common goal or a common use is the principle that unites the classes and that forms the concepts. So, the semantic relation between, for example, [food] and [fish] is in our culture normally a generic relation ([fish] are generally seen as kinds of [food]).

32 4. Theories of semantics Some groups of vegetarians may oppose this semantic relation. They may even try to get some kinds of fish (or fish in general) forbidden as food. In this way is the stability of the semantic relation between to concepts related to interests and possible social and cultural conflicts. The stability of the relation is an expression of the power relations between groups of people (in case a group of vegetarians and the rest of us).

33 4. Theories of semantics In science is the semantic relation between two term related to scientific theories and views. [Pluto] was earlier considered a [Planet] but is from 2006 no longer so considered by an official body of astronomers (it is now considered [dwarf planet]). But other astronomers disagree and the stability of this semantic relation depends on the relative strength of the two groups in the future. We may be prepared to see both views merged in the literature. In all cases are both views necessary to consider when conducting retrospective literature searches, why semantic tools should inform about the disagreement.

34 4. Theories of semantics The view of Activity Theory is that meanings, signs and documents are developed to function in relation to standardized practices in communities. We use, e.g. the Bible and the Hymn Book in our standardized religious practices. We use textbooks in our standardized teaching practices, law books in standardized legal practices etc. Concepts and documents have more or less stable functional values in relation to such standardized practices. There are of course always different views of whether such standardized practices should be changed or remain unchanged, and there are always different kinds of possible changes of those practices.

35 4. Theories of semantics Often scientific and technological development changes standardized practices in a rather uncontroversial way. In other cases, however, changes in practices are related to different political interests, to different theories or "paradigms". Different paradigms tend to influence given practices in different ways, and by doing so they also tend to change our symbolic systems as well as our production of documents and the form, content and use of those documents. The proper study of concepts, symbols, semantics and documents is thus based on the study of the functions and interests those documents are serving.

36 4a. Kinds of semantic relations Kinds of semantic relations include: Active relation: A semantic relation between two concepts, one of which expresses the performance of an operation or process affecting the other. Antonymy (A is the opposite of B; e.g. cold is the opposite of warm) Associative relation: A relation which is defined psychologically: that (some) people associate concepts (A is mentally associated with B by somebody). Causal relation: A is the cause of B. For example: Scurvy is caused by lack of vitamin C.

37 4a. Kinds of semantic relations Homonym. Two concepts, A and B, are expressed by the same symbol. Example: Both a financial institution and an edge of a river are expressed by the word bank (the word has two senses). Hyponymous relationships (hyponym-hyperonym), also termed generic relation, genus-species relation or hierarchical subordinate relation. (A is kind of B; A is subordinate to B; A is narrower than B; B is broader than A). Locative relation: A semantic relation in which a concept indicates a location of a thing designated by another concept. A is located in B; example: Minorities in Denmark.

38 4a. Kinds of semantic relations Meronymy, partitive relation (part-whole relation): a relationship between the whole and its parts (A is part of B) Passive relation: A semantic relation between two concepts, one of which is affected by or subjected to an operation or process expressed by the other. Polysemy: A polysemous (or polysemantic) word is a word that has several sub-senses which are related with one another. (A1, A2 and A3 shares the same expression) Synonymy (A denotes the same as B; A is equivalent with B). etc

39 4a. Kinds of semantic relations How many kinds of semantic relations exist? Is the number of semantic relations finite or infinite? What determines this number? Some researchers argue that there exists a small set of semantic relationships. Others argue that semantics cannot be exhausted by any finite listing of relationships. Green (2001, p. 5-6) argues that the inventory of semantic relationships includes both a closed set of relationships (including mainly hierarchical and equivalence relationships) and an open set of relationships. Every time a new verb is coined, for example, the potential for the introduction of a new conceptual relationship arises.

40 4a. Kinds of semantic relations In my own understanding must the number of semantic relations be infinite because language reflects relations among real objects and the kind of relations among real objects must be infinite. The limit to the number of semantic relations seems to be relations that nobody have found interesting to conceptualize and express. If this argument is correct then the number of semantic relations is infinite. This imply that LIS professionals cannot work from given list of semantic relations. They have partly to be uncovered in the literature to be indexed.

41 4a. Kinds of semantic relations Rosario & Hearst (2001) identified 38 semantic relations within medicine. ”We tried to produce relations that correspond to the linguistic theories such as those of Levi and Warren, but in many cases these are inappropriate.” Such findings indicates that the discovery or construction of semantic relations have to be tied to their application in KOS.

42 5. Implication for Library and Information Science (LIS) In order to discuss the relations between semantics and KO we may ask: What approaches have been used in the field of KO during its history? How do they relate to semantic theory? Broughton et al. (2005) suggested that the following traditions in KO are most important to consider:

43 5. Implication for Library and Information Science (LIS) 1.The traditional approach to KOS expressed by classification systems used in libraries and databases, including DDC, LCC and UDC 2.The facet-analytical approach founded by Ranganathan 3.The information retrieval tradition (IR) 4.User oriented / cognitive views 5.Bibliometric approaches 6.The domain analytic approach 7.Other approaches.

44 5. Implication for Library and Information Science (LIS) Given that KOS essentially are semantic tools should different approaches to KO reflect different approaches to semantics. The traditional approach to classification introduced the principle of literary warrant and thus based the semantic relations in the scientific and scholarly literature. This was (and is) often done on positivist premises: The scientific literature is seen as representing facts about knowledge and structures in knowledge and that subject specialists are able to make true and objective representations of in KO (thus tending to neglect conflicting evidence and theories).

45 5. Implication for Library and Information Science (LIS) The facet analytic approach tends to base KO more on a priory semantic relations. Its methodology is more based on the application of (logical) principles than on the study of evidence in literatures (although this is also to some degrees visible in the tradition). The IR tradition sees the semantic relations as statistical relations between signs and documents. It is atomist in the sense that it does not consider how traditions, theories and discourse communities have formed the statistical patterns it observes.

46 5. Implication for Library and Information Science (LIS) User-oriented and cognitive views tend to replace literary warrant with empirical user studies and thus to base semantic relations on users rather than on the scientific literature. The bibliometric approach considers documents to be semantically related if they cite each other, are being co-cited or bibliographic coupled. Again are the semantic relations based on some kind of literary warrant, but in a quite different way compared to the traditional approach.

47 5. Implication for Library and Information Science (LIS) The domain-analytic approach is rather traditional in its identification of semantic relations based on literary warrant. It is not positivist, however. It regards semantic relations as determined by theories and epistemologies, which more or less influence all fields of knowledge. What is indicated above is that different approaches to KO imply different views on semantics. This is, however, a point that has not been considered in the literature before.

48 5. Implication for Library and Information Science (LIS) One conclusion is that discourses about approaches to KO are intimately linked to discourses about semantic theories. Another conclusion is semantic tools should, as far as possible, uncover the different views associated with different semantic indications, e.g., that different views and theories implies different semantic relations. The determination of semantic relations within LIS should, at least ideally, involve a reflection of different views and values associated with different discourses.

49 5. Implication for Library and Information Science (LIS) This may imply that certain kinds of semantic tools, e.g. dictionaries associated with the German concept Begriffsgeschichte have been too much neglected. Of course the need for a broad spectrum for semantic tools will continue to exist, and not all tools need to be based on deep scholarly studies. However such studies are needed to gain the understanding, that is important to advance KO. Smart techniques (e.g. statistical approaches) may be able to improve overall performance of KOS, but a better scholarly understanding of meaning production and representation is needed.

50 References Eco, U. (1984). Dictionary vs. Encyclopedia (Chapter two in his: Semiotics and the Philosophy of Language. London: MacMillian. z.pdf

51 References Frohmann, B. P. (1983). An investigation of the semantic bases of some theoretical principles of classification proposed by Austin and the CRG. Cataloging & Classification Quarterly, 4(1), Green, R. (2001). Relationships in the organization of knowledge: An overview. IN: Bean, C. A. & Green, R. (Eds.). Relationships in the organization of knowledge. Dordrecht: Kluwer Academic Publishers. (Pp. 3-18).

52 References Haukioja, J. (2005). A middle position between meaning finitism and meaning Platonism. International Journal of Philosophical Studies, 13(1),

53 References Hjørland, B. (1998). Information retrieval, text composition, and semantics. Knowledge Organization, 25(1/2), Hjørland, B. (in press). Semantics and knowledge organization. Annual Review of Information Science and Technology, vol. 41 (2007).

54 References Hodge (2000). Systems of knowledge organization for Digital libraries. Beyond Traditional Authority Files. Ogden, C. K.& Richards, I. A. (1923). The Meaning of Meaning: A Study of the Influence of Language Upon Thought and of the Science of Symbolism. London: Routledge & Kegan Paul.

55 References Peirce, C. S. ( ). Collected Papers of C. S. Peirce ed. by C. Hartshorne, P. Weiss, & A. Burks, 8 vols., Harvard University Press, Cambridge, MA Rosario, B. & Hearst, M. (2004). Classifying Semantic Relations in Bioscience Text. 42nd Annual Meeting of the Association for Computational Linguistics. Forum Convention Centre Barcelona. July 21-26, relations.pdf

56 References Soergel, D. (2004). The Arts and Architecture Thesaurus (AAT). A critical appraisal.