Activating Attributes in Frames Jens Fleischhauer Department of General Linguistics Heinrich-Heine-Universität Düsseldorf Universitätsstrasse 1 40225 Düsseldorf, Germany fleischhauer@phil.uni-duesseldorf.de 1 Introduction tion is more complex than its adjectival equiva- lent. It is either possible to specify the tempo- The general topic addressed in this paper is the ac- ral extent (duration or frequency) of an eventu- tivation of scalar attributes in the context of degree ality or to specify the degree of a gradable prop- gradation of non-scalar verbs. Non-scalar verbs erty associated with the verb. The first type is such as German stinken ‘stink’ do not lexically en- called ‘extent gradation’, the second is called ‘de- code a scale, meaning there is no scalar attribute gree gradation’ (Bolinger , 1972; Löbner , 2012; in their lexical representation. Nevertheless such Fleischhauer , 2014). Two German examples of verbs can be used in a degree context as in (1). In verbal degree gradation are shown in (2). the sentence, the intensifier sehr ‘very’ specifies the intensity of the dog’s smell. (2) a. Peter ist sehr gewachsen. Peter is very grown (1) Der Hund stinkt sehr. ‘Peter has grown a lot.’ the dog stinks very b. Peter hat sehr geblutet. The dog stinks very much. Peter has very bled If the verb does not lexicalize a scale, a scalar ‘Peter bled a lot.’ attribute has to be activated in the degree context; otherwise the degree construction could not be in- In (2-a), the intensifier sehr specifies the degree terpreted. Therefore, I will argue (i) that the scalar to which Peter increased in size; it is a vague, attribute is retrieved from the conceptual knowl- context-dependent high degree (see Fleischhauer edge associated with a meaning component speci- (2013) for a deepter discussion of degree grada- fied in the verb, and (ii) that frames provide a suit- tion of change of state verbs). In (2-b) the intensi- able means of representing the process of (scalar) fier indicates the quantity of emitted blood. attribute activation. The aim of the paper is to il- There is a crucial difference between the verbs lustrate how this process is constrained. wachsen ‘grow’ and bluten ‘bleed’ in (2); the for- mer is lexically scalar, whereas the latter is not. 2 Verb gradation A verb is lexically scalar iff it expresses a scalar predication in every context of use (see, among Following Bierwisch ( 1989), gradation is a lin- others, Levin and Rappaport Hovav (2010) and guistic process of comparing two degrees on a Fleischhauer and Gamerschlag ( 2014) on scalar scale. Gradation is usually associated with ad- verbs). In (3-a) wachsen expresses a compari- jectives, and languages like English and German son between the size of the child at the beginning have special adjectival degree morphology such of the event and its size at the end of the event. as comparative -er and superlative -est in En- Hence, it expresses a scalar predication although glish. However, gradation is not restricted to ad- it is not modified by an intensifier. jectives (Sapir , 1944; Bolinger , 1972); verbs and nouns can also be graded (see e.g. Morzycki (3) a. Peter ist gewachsen. (2009) on the gradation of nouns). Verbs and Peter is grown nouns differ from adjectives in not having spe- ‘Peter has grown.’ cial degree morphemes (at least in English and b. Peter hat geblutet. German). A further difference between the gra- Peter has bled dation of adjectives and verbs is that verb grada- ‘Peter bled.’ Copyright c by the paper’s authors. Copying permitted for private and academic purposes. In Vito Pirrelli, Claudia Marzi, Marcello Ferro (eds.): Word Structure and Word Usage. Proceedings of the NetWordS Final Conference, Pisa, March 30-April 1, 2015, published at http://ceur-ws.org 58 The sentence in (3-b) does not compare the T quantity of blood emitted by the boy to some other quantity; hence, the verb is lexically non-scalar. This means that only wachsen but not bluten lexi- Object Color cally encodes a scale. COLOR : color Although the verb bluten is gradable (2-b), it Bark (...) red green blue (...) does not lexicalize a scale. The gradation scale varies for different verbs: it is an intensity scale in Figure 2: Partial type signature. (1) and a quantity scale in (2-b). Since the scale varies for different verbs, it is not contributed by a non-scalar attribute, such as COLOR. the intensifier. Rather, a suitable gradation scale is To restrict the admissible attributes for a frame rather from conceptual knowledge. and the admissible values for an attribute, types 3 Frames can be assigned to frames. Types are ordered with regard to their specificity in a type signa- Frames, in the sense of Barsalou ( 1992a; 1992b), ture (Carpenter , 1992), as shown in figure 2. The are recursive attribute-value structures. A frame type signature defines ‘bark’ as a subtype of the is a representation of a concept and represents the type ‘object’; ‘red’, ‘green’ and ‘blue’ are de- referent of the concept in terms of its attributes, the fined as subtypes of ‘color’. The type signature values of the attributes, the attributes of the values is enriched with appropriateness conditions (ACs) and so on. One way of representing frames is by which serve two tasks: first, they restrict the set using attribute-value matrixes (AVMs) like in fig- of appropriate attributes for frames to a certain ure 1. The AVM in figure 1 shows a partial frame type. Second, ACs specify the appropriate values for the concept ‘tree’ (based on Petersen and Os- for an attribute; it is required that all values of an swald (2012)). A tree consists of a crown and a attribute are of a certain type (see Petersen (2007), 1 trunk, hence CROWN and TRUNK are attributes Petersen et al. (2008), Petersen and Gamerschlag in the frame of ‘tree’. The value of the attribute (2014)). COLOR restricts its values to be of the CROWN is the underspecified value or, in different type ‘color’ or one of its subtypes. Furthermore, terms, the uninstantiated type ‘crown’. The value the attribute COLOR is an appropriate attribute for of trunk is the uninstantiated type ‘trunk’ which ‘object’. Since ‘bark’ is a subtype of ‘object’, it can be further characterized as having an attribute inherits this AC. Thus, objects of the type ‘bark’ BARK . The bark of the tree is characterized as hav- have a color but do not have, for example, a price, ing a certain color. since the type signature does not define PRICE as   an appropriate attribute for ‘bark’. tree   4 Frame analysis of degree gradation CROWN crown      h i   In section 2, I suggested that the degree context TRUNK trunk BARK bark COLOR color activates the relevant gradation scale in the case of lexically non-scalar verbs. This process is not Figure 1: Partial frame for the concept ‘tree’. arbitrary but restricted by the lexical semantics of the verb. There are two reasons for this assump- Following Löbner (1998; 2014) and Petersen tion: First, each semantic class of gradable verbs (2007), attributes are partial functions; they as- is only related to a single gradation scale. Sec- sign a unique value to their possessor argument. ond, different semantic classes of verbs are related The requirement of functionality provides a for- to different gradation scales. As discussed above, mal constraint on possible attributes. As attributes verbs of substance emission such as bluten ‘bleed’ are functions, it is possible to distinguish scalar are related to a quantity scale (2-b), but verbs of and non-scalar attributes by looking at their do- smell emission, like stinken ‘stink’ in (1), are re- mains. If the values in the domain are linearly or- lated to an intensity scale. dered, the attribute is a scalar one (e.g. SIZE). If In the following, the analysis concentrates on there is no linear order of the domain’s values, it is the verb bluten. The verb denotes a process of sub- 1 Attributes are written in small capitals. stance emission. Its single argument is the emit- 59 ter, the one who is emitting blood. The emit- only attribute that can be activated in a degree con- tee, which is the emitted substance, is an im- text to provide a suitable gradation scale. plicit semantic argument of the verb (Goldberg I propose the constraint in (4) as a restriction for (2005) speaks of an incorporated theme argu- the activation of scalar attributes in the frames of ment). A frame representation for bluten, cap- lexically non-scalar verbs: turing the mentioned aspects, is given in figure 3. The boxed numeral in the frame indicates structure (4) Only meaning components that are lexi- sharing (Pollard and Sag , 1994) and indicates that cally specified in the verb license the ac- the value of EMITTER is coextensive with a some tivation of scalar attributes. other structure, the externally specified subject. In the frame for bluten (figure 3) only the emittee   is lexically specified as being blood. The emitter is substance emission   not specified in the verb, rather it is introduced by EMITTER 1    the subject argument and therefore does not give EMITTEE blood access to specific conceptual knowledge. Figure 3: Frame for the verbal concept bluten 5 Restricting the scalar attribute ‘bleed’. An apparent problem is the claim that the frame Degree gradation affects the quantity of the for bluten only contains one scalar attribute, emitted blood; hence QUANTITY is an attribute namely QUANTITY. It is clearly the case that we of the emittee. The frame representation for sehr cannot only speak of the quantity of blood but also bluten ‘bleed a lot’ is shown in fig 4. The inten- of its temperature or pressure. TEMPERATURE as sifier sehr activates the scalar attribute QUANTITY well as PRESSURE are scalar attributes too, so the in the frame of bluten and specifies the value of question emerges why it is only QUANTITY but not QUANTITY as ‘high’. TEMPERATURE or PRESSURE that is activated in a   degree context? substance emission To tackle this problem one has to realize that   EMITTER 1  the gradable verbs of substance emission are not  h i   restricted to those that express an emission of a EMITTEE blood QUANTITY high liquid like blood. Other verbs of this class express the emission of a solid like hair in (5). Figure 4: Frame for sehr bluten ‘bleed a lot’. (5) Die Katze hat sehr gehaart. As QUANTITY is an attribute of ‘blood’, it is the cat has very shed the object knowledge associated with ‘blood’ that ‘The cat lost many hairs.’ licenses its activation. A partial frame for ‘blood’ is given in figure 5. The type signature in figure 6 defines ‘liquid’   to be a supertype of ‘blood’ and ‘water’ and to be blood a subtype of ‘substance’. ‘Solids’ are also a sub-   type of ‘substance’ and form the supertype of, for CONSISTENCY liquid    COLOR red  example, ‘hair’ and ‘scall’. The attributes shared   by liquids and solids are inherited from their com- QUANTITY quantity mon supertype, for example CONSISTENCY and Figure 5: Partial frame for the concept Blut QUANTITY . But there are attributes which ‘hair’ ‘blood’. and ‘blood’ do not share and these are inherited from the more specific supertypes ‘liquids’ and It is part of our knowledge of ‘blood’ that it has ‘solids’ respectively. For example, liquids do have a certain consistency (‘liquid’), has a certain color a temperature and a pressure but we do not think (‘red’) and is of a certain quantity. While the at- of solids in terms of the attributes PRESSURE and tributes CONSISTENCY and COLOR have fixed val- TEMPERATURE. This does not result in the claim ues for blood, the value of QUANTITY is depen- that solids do not have a temperature but I do not dent on the possessor of the blood. In figure 5 the think that TEMPERATURE is an attribute in our ob- only scalar attribute is QUANTITY, hence it is the ject knowledge of ‘hair’ or ‘scall’; so we do not 60 T tribute is activated from the conceptual knowledge associated with a meaning component lexically Substance specified in the verb. Furthermore, the gradable at- CONSISTENCY : consistency tributes that can be activated are restricted to those QUANTITY : quantity inherited from the most specific common super- type. This ensures a homogeneous interpretation of degree gradation of verbs of substance emis- Liquids Solids sion, otherwise degree gradation of verbs (of sub- CONSISTENCY : liquid CONSIST.: solid stance emission) would be totally idiosyncratic. TEMPERATURE: temperature Hair Scall Frames provide a suitable framework for the PRESSURE: pressure analysis of the sketched phenomenon as they al- Blood Water low representing lexical knowledge and concep- tual knowledge in the same representational for- Figure 6: Partial type signature. mat. The frame analysis in this paper concentrates on a single semantic verb class but it can easily be extended to cover other classes of gradable verbs, represents these concepts by using the attribute for example verbs of smell/light/sound emission TEMPERATURE. or experiencer verbs, too. I propose that the gen- As verbs of substance emission do not only ex- eral constraints formulated in (4) and (6) hold for press the emission of liquids but of solids too, the these classes of verbs as well, the only difference admissible scalar attributes that can be activated consists in the associated conceptual knowledge. in a degree context are restricted to those inherited The process of attribute activation is not re- from the common supertype of liquids and solids, stricted to scalar attributs in the context of ver- which is ‘substance’. Since QUANTITY but not bal degree gradation. A similar process occurs if TEMPERATURE or PRESSURE is inherited from verbs of sound emission are used for denoting mo- ‘substance’, it is only QUANTITY that can be ac- tion events like in (7) (based on Kaufmann (1995, tivated in the context of degree gradation. Beside 93)). In this construction, a motion frame is acti- the constraint in (4) a further constraint restricting vated which is licensed by the fact that the motion the activation of scalar attributes is required: of a motorbike produces a yowling sound. In this (6) The activation of scalar attributes is re- case and in opposition to verbal degree gradation, stricted to those attributes which are inher- knowledge of the subject referent is relevant too. ited form the most specific common super- type. (7) Das Motorrad jaulte über die the motorbike yowled over the The most specific common supertype for emit- Kreuzung. table substances like ‘blood’ and ‘hair’ is ‘sub- crossing stance’. Hence, (6) restricts the activation of scalar ‘The motorbike yowled over the crossing.’ attributes to those which are inherited from ‘sub- stance’; those attributes inherited from a more spe- It is a promising task for the future to explore cific supertype like ‘liquids’ cannot activated in a the process of attribute activation in more details degree context. and to see how the activation of attributes from the conceptual knowledge is constrained by lexical se- 6 Conclusion mantics and other factors. In this paper, I have shown that lexically non- scalar verbs can be graded by intensifiers like sehr. Acknowledgments But this requires the activation of a suitable scalar attribute, otherwise the degree construction could The paper is a result of my work in the Collabora- not be interpreted. The process of attribute activa- tive Research Center “The Structure of Represen- tion is not unconstrained, rather the lexical mean- tations in Language, Cognition, and Science” sup- ing of the verb as well as conceptual knowledge ported by the German Science Foundation (DFG). provide constraints on this process. The scalar at- 61 References [Löbner 2014] Sebastian Löbner. 2014. Evidence for frames from natural language. In T. Gamerschlag, [Barsalou 1992a] Lawrence Barsalou. 1992a. Cogni- D. Gerland, R. Osswald, and W. Petersen (eds.), tive Psychology. An overview for cognitive scientists. Frames and Concept Types: Applications in Lan- Hillsdale/ NJ, Lawrence Erlbaum Association. guage and Philosophy, 23–68. Dordrecht/ Heidel- berg/ New York, Springer. [Barsalou 1992b] Lawrence Barsalou. 1992b. Frames, concepts, and fields. In A. Lehrer and E. F. Kittay [Morzycki 2009] Marcin Morzycki. 2009. Degree (eds.), Frames, fields, and contrasts, 21–74. Hills- modification of gradable nouns: size adjectives and dale/ NJ, Lawrence Erlbaum Association. adnominal degree morphemes. Natural Language Semantics 17:175–203. [Bierwisch 1989] Manfred Bierwisch. 1989. The Se- mantics of Gradation. In M. Bierwisch and E. Lang [Petersen 2007] Wiebke Petersen. 2007. Representa- (eds.), Dimensional Adjectives, 71–261. Berlin, tion of Concepts as Frames. In J. Skilters, F. Toc- Springer. cafondi, and G. Stemberger (eds.), Complex Cog- nition and Qualitative Science. The Baltic Interna- [Bolinger 1972] Dwight Bolinger. 1972. Degree tional Yearbook of Cognition, Logic and Communi- Words. Mouton, The Hague. cation. Vol 2, 151–170. Riga, University of Latvia. [Petersen et al.2008] Wiebke Petersen, Jens Fleis- [Carpenter 1992] Bob Carpenter. 1992. The Logic of chhauer, Peter Bücker, and Hakan Beseoglu. 2008. Typed Feature Structures. Cambridge, Cambridge A Frame-based Analysis of Synaesthetic Metaphors. University Press. The Baltic International Yearbook of Cognition, Logic and Communication. Vol 3, 1–22. [Fleischhauer 2013] Jens Fleischhauer. 2013. Interac- tion of telicity and degree gradation in change of [Petersen and Gamerschlag 2014] Wiebke Petersen and state verbs. In B. Asrenijevic, B. Gehrke and R. Thomas Gamerschlag. 2014. Why chocolate eggs Marin (eds.), Studies in Composition and Decom- can taste old but not oval: A frame-theoretic anal- position of Event Predicates, 125–152. Dordrecht, ysis of inferential evidentials. In T. Gamerschlag, Springer. D. Gerland, R. Osswald, and W. Petersen (eds.), Frames and Concept Types: Applications in Lan- [Fleischhauer 2014] Jens Fleischhauer. 2014. Degree guage and Philosophy, 199–220. Dordrecht/ Hei- Gradation of Verbs. Doctoral dissertation, Heinrich- delberg/ New York, Springer. Heine-Universität Düsseldorf. [Petersen and Osswald 2012] Wiebke Petersen and [Fleischhauer and Gamerschlag 2014] Jens Fleis- Tanja Osswald. 2012. A Formal Interpretation of chhauer and Thomas Gamerschlag. 2014. We’re Concept Types and Type Shifts. In K. Kosecki and going through changes: How change of state verbs J. Badio (eds.), Cognitive Processes in Language, an arguments combine in scale composition. Lingua 183–191. Frankfurt, Peter Lang. 141:30–47. [Pollard and Sag 1994] Carl Pollard and Ivan A. Sag. 1994. Head-Driven Phrase Structure Grammar. [Goldberg 2005] Adele Goldberg. 2005. Argument re- Chicago, The University of Chicago Press. alization: The Role of constructions, lexical seman- tics and discourse factors. In J.-O. Ostman and M. [Rappaport Hovav and Levin 2010] Malka Rappaport Fried (eds.), Construction Grammars: Cognitive Hovav and Beth Levin. 2010. Reflections on Grounding and Theoretical Extensions, 17–43. Am- manner/result complementarity. In M. Rappaport sterdam, John Benjamins. Hovav, E. Doron and I. Sichel (eds.), 21–38. Oxford, Oxford University Press. Syntax, Lexical [Kaufmann 1995] Ingrid Kaufmann. 1995. What is an Semantics, and Event Structure (im)possible verb? Restrictions on Semantic Form and their consequences for argument structure. Fo- [Sapir 1944] Edward Sapir. 1944. Grading: A Study in lia Linguistica XXIX/1-2:67–103. Semantics. Philosophy of Science 11(2):93–116. [Löbner 1998] Sebastian Löbner. 1998. Definite Asso- ciative Anaphora. In S. Botley (ed.), Proceedings of DAARC96 - Discourse Anaphora and Resolution Colloquium. Lancaster University, July 17th-18th. Lancaster. [Löbner 2012] Sebastian Löbner. 2012. Sub- compositionality. In M. Werning, W. Hinzen and E. Machery (eds.), The Oxford Handbook of Com- positionality, 220–241. Oxford, Oxford University Press. 62