Decompositional Semantics Rachel Rudinger January 30, 2020 A story - PowerPoint PPT Presentation

Decompositional Semantics Rachel Rudinger January 30, 2020

A story about semantic annotation…

Traditional Semantic Annotation Who did what to whom? AGENT PATIENT Alex shattered the window. Participant that performs the action. AGENT Participant that undergoes the action. PATIENT

Traditional Semantic Annotation ??? AGENT PATIENT Alex shattered the window with a hammer. Participant that performs the action. AGENT Participant that undergoes the action PATIENT and changes state.

Traditional Semantic Annotation AGENT PATIENT INSTRUMENT Alex shattered the window with a hammer. Participant that performs the action. AGENT Participant that undergoes the action PATIENT and changes state. INSTRUMENT Participant used to carry out the action.

Traditional Semantic Annotation ??? PATIENT The cold air shattered the window. Participant that performs the action. AGENT Participant that undergoes the action PATIENT and changes state. INSTRUMENT Participant used to carry out the action.

Traditional Semantic Annotation FORCE PATIENT The cold air shattered the window. Participant that performs the action Participant that causes the action AGENT FORCE with intent. without intent. Participant that undergoes the action PATIENT and changes state. INSTRUMENT Participant used to carry out the action.

Traditional Semantic Annotation AGENT ? FORCE ? ??? PATIENT Alex accidentally shattered the window. Participant that performs the action Participant that causes the action AGENT FORCE with intent. without intent. Participant that undergoes the action PATIENT and changes state. INSTRUMENT Participant used to carry out the action.

Traditional Semantic Annotation FORCE ? I NSTRUMENT ? ??? PATIENT Alex’s singing shattered the window. Participant that performs the action Participant that causes the action FORCE AGENT with intent. without intent. Participant that undergoes the action PATIENT and changes state. INSTRUMENT Participant used to carry out the action.

VerbNet Role Hierarchy A hierarchical unification of LIRICS and VerbNet semantic roles . Bonial, Corvey, Palmer, Petukhova, and Bunt. ICSC. 2011.

Does this fall Does this fall Practical Challenges into category into any A or B? category? Annotation Establish challenges. ontology. Modify ontology. Train expert Retrain? annotators. Re-annotate? Mapping Annotate. between ontologies?

Dowty (1991) “…and as soon as we try to be precise about exactly what Agent, Patient, etc., ‘mean’, it is all too subject to difficulties and apparent counterexamples.” “…we may have a hard time pinning down the traditional role type because role types are simply not discrete categories at all, but rather are cluster concepts” Thematic proto-roles and argument selection. David Dowty. Language. 1991.

Dowty’s Proto-Agent and Proto-Patient Properties (“ (“Semantic Proto-Ro Roles”) Thematic proto-roles and argument selection. David Dowty. Language. 1991.

The Decompositional Approach Identify Instigated Instigated Extend Awareness properties Awareness inventory of Physical of interest. Physical … properties. Sentient Moved Destroyed Translate … properties into Did ARG cause the PRED templatic English to happen? questions. Pose each Did ARG question Make new change location during PRED? independently to annotations (but non-expert keep the old)! annotators.

Decompositional Semantics Initiative “Rapid, simple, commonsensical annotations of meaning” 1. Target aspects of meaning at the phrase- or sentence level. 2. Simple, linguistically- or cognitively-motivated properties. 3. Many independent labels. 4. Straightforward questions for crowd workers. http://decomp.io

Decompositional Semantics Initiative “Rapid, simple, commonsensical annotations of meaning” Semantic Proto-Roles Genericity Time Event Factuality Decomp Diverse Natural PredPatt Word Sense Toolkit Language Inference Cross-lingual Decompositional Common Sense ParaBank 1 & 2 Semantic Parsing Inference http://decomp.io

Dataset 1: Semantic Proto-Roles Dataset 2: Event Factuality Dataset 3: Temporal Relations Dataset 4: Genericity

Before we dive into the data…

Predicate-Argument Identification with P RED P ATT • Decomp annotation protocols rely on predicate-argument structure. • PredPatt: series of rules to map Universal Dependencies (UD) parse to unlabeled predicate-argument structure. • Scalability and (potential) Multilinguality: Piggy-backing on UD resources. ?a extracts ?b from ?c ?a: PredPatt ?b: predicates ?c: text ?a extracts ?b from ?c ?a: PredPatt ?b: arguments ?c: text

loves loves nsubj dobj Chris Pat Chris Pat slide courtesy Aaron Steven White, 2019

loves loves nsubj dobj human greyhound human greyhound det det the the slide courtesy Aaron Steven White, 2019

told told nsubj dobj ccomp Chris Pat SOMETHING Chris Pat built nsubj dobj built boy boat det det boat boy a a slide courtesy Aaron Steven White, 2019

Important note No typing beyond: - event v. participant - argument v. head slide courtesy Aaron Steven White, 2019

Hiller asked Bush to name the leaders of Chechnya , Taiwan , India and Pakistan slide courtesy Aaron Steven White, 2019

Hiller asked Bush to name the leaders of Chechnya , Taiwan , India and Pakistan syntax node syntax edge slide courtesy Aaron Steven White, 2019

relation event participant Hiller asked Bush to name the leaders of Chechnya , Taiwan , India and Pakistan arg. edge semantic head edge predicate node argument node syntax node syntax edge slide courtesy Aaron Steven White, 2019

event participant Hiller asked Bush to name the leaders of Chechnya , Taiwan , India and Pakistan arg. edge semantic head edge predicate node argument node syntax node syntax edge instance edge nonhead edge slide courtesy Aaron Steven White, 2019

Hiller asked Bush to name the leaders of Chechnya , Taiwan , India and Pakistan arg. edge semantic head edge predicate node argument node syntax node syntax edge instance edge nonhead edge slide courtesy Aaron Steven White, 2019

Subspace Attribute Val factuality factual 1.038 genericty pred-dynamic 1.418 Subspace Attribute Val genericity pred-hypothetical -0.892 protoroles awareness -0.110 Subspace Attribute Val genericity 1.418 pred-particular protoroles change-of-loc -0.039 time dur-days -1.062 genericity arg-abstract -1.112 protoroles change-of-poss 0.000 time dur-minutes -0.912 genericity arg-kind 1.195 protoroles change-of-state -0.104 time dur-seconds 1.260 genericity arg-particular -1.112 protoroles existed-before 1.402 … word-sense noun.act -3.000 … word-sense noun.cognition -3.000 word-sense noun.food -3.000 … Hiller asked Bush to name the leaders of Chechnya , Taiwan , India and Pakistan arg. edge semantic head edge predicate node argument node syntax node syntax edge instance edge nonhead edge slide courtesy Aaron Steven White, 2019

Diving into the data…

Dataset t 1: Sem emanti tic Proto-Ro Roles Dataset 2: Event Factuality Dataset 3: Temporal Relations Dataset 4: Genericity

Traditional Semantic Role Labeling AGENT PATIENT INSTRUMENT Alex shattered the window with a hammer. Participant that causes the action FORCE Participant that performs the action. AGENT without intent. Participant that undergoes the action PATIENT and changes state. Etc… INSTRUMENT Participant used to carry out the action.

Dowty (1991) “…and as soon as we try to be precise about exactly what Agent, Patient, etc., ‘mean’, it is all too subject to difficulties and apparent counterexamples.” “…we may have a hard time pinning down the traditional role type because role types are simply not discrete categories at all, but rather are cluster concepts” Thematic proto-roles and argument selection. David Dowty. Language. 1991.

Dowty’s Proto-Agent and Proto-Patient Properties (“ (“Semantic Proto-Ro Roles”) Thematic proto-roles and argument selection. David Dowty. Language. 1991.

The Decompositional Approach Identify Instigated Instigated Extend Awareness properties Awareness inventory of Physical of interest. Physical … properties. Sentient Moved Destroyed Translate … properties into Did ARG cause the PRED templatic English to happen? questions. Pose each Did ARG question Make new change location during PRED? independently to annotations (but non-expert keep the old)! annotators.

Semantic Proto-Role Properties STATIONARY INSTIGATION CREATED LOCATION VOLITION DESTROYED PHYSICAL CONTACT CHANGED AWARENESS MANIPULATED CHANGED STATE SENTIENT WAS USED CHANGED POSSESSION PHYSICALLY EXISTED PARTITIVE CHANGED LOCATION EXISTED BEFORE … AND MORE ? CHANGED STATE CONTINUOUS EXISTED DURING WAS FOR BENEFIT EXISTED AFTER Semantic Proto-Roles. Reisinger, Rudinger, Ferraro, Harman, Rawlins, and Van Durme. TACL. 2015.