The interplay between conceptual and referential aspects of meaning - PowerPoint PPT Presentation

The interplay between conceptual and referential aspects of meaning Gemma Boleda Universitat Pompeu Fabra (work in collaboration with Louise McNally) BRIDGE Workshop ESSLLI 2018, 6–10 August 2018, Sofia, Bulgaria 1

Acknowledgements ◮ Co-authors in cited papers ◮ Laura Aina, Kristina Gulordava, Carina Silberer, Ionut-Teodor Sorodoc, Matthijs Westera ◮ This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 715154). AMORE: A distributional Model Of Reference to Entities). 2

The problem Modifier-noun relations bifurcate: ◮ Strong default interpretations ◮ In context, anything goes How to explain this? 3

Example: adjectives with little/no context Canadian visit / attack / decision . . . must denote agent (Kayne 1984, a.o.) 4

Example: adjectives with little/no context Canadian visit / attack / decision . . . must denote agent (Kayne 1984, a.o.) (1) Yeltsin met the prospective Democratic presidential candidate Bill Clinton on June 18. His itinerary also included an official visit to Canada 4

Example: adjectives with little/no context Canadian visit / attack / decision . . . must denote agent (Kayne 1984, a.o.) (1) Yeltsin met the prospective Democratic presidential candidate Bill Clinton on June 18. His itinerary also included an official visit to Canada /?? an official Canadian visit . 4

Example: adjectives with little/no context Canadian visit / attack / decision . . . must denote agent (Kayne 1984, a.o.) (1) Yeltsin met the prospective Democratic presidential candidate Bill Clinton on June 18. His itinerary also included an official visit to Canada /?? an official Canadian visit . (2) Put the scarf in the red box . 4

Adjectives in context (3) Prince Edward and wife begin Canadian visit 5

Adjectives in context (3) Prince Edward and wife begin Canadian visit (4) (Context: For a fundraising sale, Adam and Barbara are sorting donated scarves according to color in different, identical, brown cardboard boxes. Barbara distractedly puts a red scarf in the box containing blue scarves.) Adam: Hey, this one belongs in the red box ! 5

More specific questions ◮ Strong default interpretations ◮ Why does the default seem so strong? ◮ In context, anything goes ◮ Why/How can context ameliorate anything? ◮ What kind of theory can account for this compositional phenomenon? 6

Previous work: Two general approaches to modification ◮ Semantic primitives ◮ Underspecification of modification relation + resolution in context The closest thing we have seen to a mixed approach appears in Asher (2011). 7

Resolution via semantic primitives ◮ Long tradition ◮ Makes explicit how the concepts introduced by the modifier and the head are composed ◮ Examples: ◮ Levi (1978): CAUSE, HAVE, MAKE, USE, BE, IN, FOR, FROM, ABOUT, ACT, PRODUCT, AGENT, PATIENT ◮ Pustejovsky (1995): FORMAL, CONSTITUTIVE, AGENTIVE, TELIC ◮ Ó Séaghdha and Copestake (2009): BE, HAVE, IN, AGENT, INSTRUMENT, ABOUT 8

Resolution via semantic primitives (5) Canadian visit : λ x . visit ( x ) ∧ AGENT ( x , Canada ) (6) red apple : λ x ∃ y . apple ( x ) ∧ CONSTITUTIVE ( apple )= PART-OF (y,x) ∧ red (y) 9

Pros and cons ◮ Pros ◮ intuitions about default interpretations ◮ predicts productivity ◮ Cons ◮ too strong ◮ too weak ◮ huge methodological issues 10

Underspecification + context ◮ Also a long tradition ◮ Relations are established indexically or by valuing a variable that stands for the relation ◮ Examples: ◮ Bosch (1983), Rothschild and Segal (2009): Adjectives denote functions from contexts to contents ◮ McNally and Boleda (2004), Kennedy and McNally (2010): Adjectives introduce variables over relations that are valued by context 11

Underspecification + context (7) Canadian visit : λ x . visit ( x ) ∧ R i ( x , Canada ) (compare to λ x . visit ( x ) ∧ AGENT ( x , Canada ) 12

Pros and cons ◮ Pro: appropriately flexible ◮ Con: too weak ◮ Pending: a theory of how context plays its role 13

Summing up default context-dependent semantic primitives ( ✔ ) ✘ underspecification + context ( ✔ ) ✘ 14

Synthesis: Conceptual and referential affordance in language McNally and Boleda 2017 Distinct aspects of language afford concept composition in different ways: ◮ The concepts described ◮ The entities referred to Affordance (Chemero (2003), based on Gibson (1979)): ◮ relation between features of situations and abilities of organisms 15

Synthesis: Conceptual and referential affordance in language Assumption (semiotic models, a.o.): Proposal ◮ the connection to concepts and to the world are distinct features of language ◮ each of them affords distinct composition process ◮ speakers avail themselves of both 16

Conceptual affordance The concepts contributed by the components of a phrase suggest the ways in which they should be composed 17

Conceptual affordance The concepts contributed by the components of a phrase suggest the ways in which they should be composed → default interpretations, little or no need for context 17

Conceptual affordance The concepts contributed by the components of a phrase suggest the ways in which they should be composed → default interpretations, little or no need for context ◮ productive: speakers use regularities in our lexical knowledge 17

Conceptual affordance The concepts contributed by the components of a phrase suggest the ways in which they should be composed → default interpretations, little or no need for context ◮ productive: speakers use regularities in our lexical knowledge (8) Put the scarf in the red box . 17

Referential affordance Independently available information about the referent indicates how the concepts should be composed 18

Referential affordance Independently available information about the referent indicates how the concepts should be composed → Ad hoc interpretations, heavy context dependence 18

Referential affordance Independently available information about the referent indicates how the concepts should be composed → Ad hoc interpretations, heavy context dependence ◮ plastic: speakers use information about the world 18

Referential affordance Independently available information about the referent indicates how the concepts should be composed → Ad hoc interpretations, heavy context dependence ◮ plastic: speakers use information about the world (9) (Context: For a fundraising sale, Adam and Barbara are sorting donated scarves according to color in different, identical, brown cardboard boxes. Barbara distractedly puts a red scarf in the box containing blue scarves.) Adam: Hey, this one belongs in the red box ! 18

Conceptual vs. referential effects in composition Asher 2011, McNally and Boleda 2017 “cafetera italiana” ( Italian coffee maker ) 19

Conceptual vs. referential effects in composition Asher 2011, McNally and Boleda 2017 “cafetera italiana” ( Italian coffee maker ) ⇓ conceptually afforded composition 19

Conceptual vs. referential effects in composition Asher 2011, McNally and Boleda 2017 “cafetera italiana” “cafetera italiana”, too! ( Italian coffee maker ) ( Italian coffee maker ) ⇓ conceptually afforded composition 19

Conceptual vs. referential effects in composition Asher 2011, McNally and Boleda 2017 “cafetera italiana” “cafetera italiana”, too! ( Italian coffee maker ) ( Italian coffee maker ) ⇓ ⇓ conceptually afforded referentially afforded composition composition 19

Contribution 1 ◮ Strong default interpretations ◮ Why does the default seem so strong? → conceptually afforded modification → model with distributional semantics ◮ In context, anything goes ◮ Why/How can context ameliorate anything? ◮ What kind of theory can account for this compositional phenomenon? 20

Distributional semantics for conceptually afforded modification ◮ default interpretations are very sensitive to the lexical semantics of the phrase components ◮ both coarse- and fine-grained default semantic primitives ( ✔ ) underspecification + context ✘ ◮ distributional semantics provides the necessary information, like primitive-based accounts, without their drawbacks 21

Distributional semantics Aka vector-space semantics, related to Neural Networks / deep learning (See Stefan Evert’s course this week at ESSLLI for more!) 22

Distributional semantics Aka vector-space semantics, related to Neural Networks / deep learning (See Stefan Evert’s course this week at ESSLLI for more!) likely) mug of bourbon in hand. Some stewed milk into a heavy mug, granules of holding his coffee mug cupped in his hands. drained his mug, dropping it over his tablespoons of coffee and a single mug of milk into the mug plus four spoons of sugar placing the empty mug on the floor picking up my mug with one hand and followed by a very hot mug of tea into which from time to time to drink a mug of tea. The briefed, relax over a mug of tea and a cake and cheese and a mug of strong, black then we had a mug of cocoa and a gingerbread and a white mug with a blurred inscription. was carrying a mug of tea and 22