Semantic Role

A Participant is an Edge that specifies the semantic role of an entity involved in an event within a predication.

Event Node (verb)
    ├─ PARTICIPANT Edge (role=Agent) ──→ Entity Node
    ├─ PARTICIPANT Edge (role=Theme) ──→ Entity Node
    └─ PARTICIPANT Edge (role=Instrument) ──→ Entity Node

Design Principles

Separation Principle

Category	Belongs to	Examples
Participants	Event level	Agent, Theme, Recipient
Pragmatic info	Context/Claim level	Speaker, Listener, Evidentiality

Speaker, Listener, and Source (information origin) are not participants but are handled in Qualifier or Context/Claim.

Encoding

4 bits (0x0~0xF), up to 16 semantic roles
Pattern matching via SIMD bit operations

Semantic Role List (16)

Core Participants

ID	Code	Role	Definition	Example
0x0	AGT	Agent	The entity that intentionally performs an action	“John kicked the ball”
0x1	EXP	Experiencer	The entity that experiences emotion/cognition/perception	“Mary was sad”
0x2	THM	Theme	The entity that moves or whose state is described	“John kicked the ball”
0x3	PAT	Patient	The entity whose state changes due to an action	“The window broke”
0x4	RCP	Recipient	The entity that receives something	“He gave the book to Mary”
0x5	BNF	Beneficiary	The entity that benefits from an action	“Made it for the child”

Instruments & Means

ID	Code	Role	Definition	Example
0x6	INS	Instrument	The tool used to perform an action	“Drove the nail with a hammer”
0x7	MNR	Manner	The way an action is performed	“Ran quickly”

Spatial

ID	Code	Role	Definition	Example
0x8	LOC	Location	The place where an event occurs	“Lived in Seoul”
0x9	SRC	Source	The starting point of movement	“Departed from home”
0xA	DST	Destination	The ending point of movement	“Went to school”
0xB	PTH	Path	The waypoint of movement	“Went through the park”

Causal

ID	Code	Role	Definition	Example
0xC	CAU	Cause	The cause of an event	“Cancelled because of rain”
0xD	PRP	Purpose	The purpose of an action	“Went to exercise”

Others

ID	Code	Role	Definition	Example
0xE	COM	Comitative	The accompanying entity	“Went with a friend”
0xF	ATR	Attribute	State/attribute predication	“The sky is blue”

Participant Edge Structure

PARTICIPANT Edge {
    source:     Event SIDX       // verb node
    target:     Entity SIDX      // entity node
    role:       4-bit            // semantic role (0x0~0xF)
    gram_role:  2-bit (optional) // grammatical role (subject/object/complement)
    focus:      4-bit (optional) // emphasis level (0~15 → 0.0~1.0)
    quant_ref:  TID (optional)   // quantifier reference
}

Field	Bits	Description
role	4	Semantic role (required)
gram_role	2	0=unspecified, 1=subject, 2=object, 3=complement
focus	4	Informational importance (0=background, 15=core emphasis)
quant_ref	16	Quantifier TID for “every”, “most”, etc.

Theme vs Patient

Role	State change	Example
Theme	None (movement/description)	“threw the ball” (ball unchanged)
Patient	Yes (affected)	“broke the glass” (glass state changed)

In practice, they can be merged as Theme and distinguished by verb semantics when needed.

Examples

Simple sentence: “John gave Mary a book”

Event: give.v.01
├─ PARTICIPANT (AGT) → John
├─ PARTICIPANT (THM) → book
└─ PARTICIPANT (RCP) → Mary

Complex sentence: “Because of rain, I ran quickly from home to school with a friend”

Event: run.v.01
├─ PARTICIPANT (AGT) → [speaker]
├─ PARTICIPANT (CAU) → rain
├─ PARTICIPANT (COM) → friend
├─ PARTICIPANT (SRC) → home
├─ PARTICIPANT (DST) → school
└─ PARTICIPANT (MNR) → quickly

State predication: “The sky is very blue”

Event: be.v.01
├─ PARTICIPANT (THM) → sky
└─ PARTICIPANT (ATR) → blue (focus=15)

Active/Passive Normalization

Surface form	Agent	Theme
“Apple acquired Tesla”	Apple	Tesla
“Tesla was acquired by Apple”	Apple	Tesla

Normalized to the same pattern during the parsing stage.