Linguistic Annotation FrameworkSC4 WG 1

Nancy Ide • Vassar College • USA

LAF Goal

ISO TC37 SC4 - WG 1

Provide a generic means to represent linguistic data and annotations Based on a formal model Users map their formats into/out of LAF

User formats must conform to underlying model

Pivot or “dump” format for exchange, machine processing

User A’s representation

User B’s representation

DUMP FORMAT

“interlingua”

Principles Separation of data and annotations

Stand-off annotation Separation of user annotation formats and

the exchange (“dump”) format Mappable to one another

Separation of referential structure and annotation content in dump format

Separation of annotation structure (relationships among parts) and content (data categories) in representation of annotations

LAF Development LAF has gone through a slow evolution

Model development (GMT as base) Consideration of processing needs Application to different annotation

types/structures/formats Adjustments to development in other WGs on

specific annotation types and feature structures “Proof of concept” instantiation in the American

National Corpus Transduction of several different annotation types

and formats to LAF format API to merge, transduce to other formats

LAF Status Have now

Reduced FS specification Final XML format / schema

GrAF : Graph Annotation Format Mapping “rules” and examples

Also Coordination with UIMA Header specification including information about

annotation, similar to UIMA type definition

Basic Model Annotation content represented by

feature structures Powerful means to represent any/all

annotations Referential structure represented as a

directed acyclic graph (DAG) Enables exploitation of well-understood

graph traversal and manipulation algorithms

Referential Structure Means by which annotation content is

associated with primary data or other annotations Very simple DAG model No need to consider internal structure of

annotation content (i.e. relations among bits of annotation information)

Primary Data Primary data contains no annotations

“Read-only” Modifications can be regarded as

annotations Insistence on the identification of a

base segmentation of the primary data Identifies contiguous sequences of

indivisible logical units For text, usually a character

“Compatible” annotations (i.e. those that can be merged etc.) use common base segmentation

Primary Segmentation Set of disjoint edges over primary data

Vertices Virtual, located between each logical unit Sequentially numbered

EdgesEach edge (x,y) in the graph delimits a non-

divisible region of primary dataComformance to MAF, SynAF call these edges over primary data a span

Multiple primary segmentations may be defined over a single primary data set Specify segmentations at different levels of

granularity A segmentation is “primary” vis a vis a given

annotation, not the data itself Edges in a primary segmentation can be

defined over any span of contiguous primary data, regardless of its length

No need for spans to be contiguous For text, most common primary

segmentation is the token

Referring to Primary Segmentation Define an edge graph over the edges

(spans) in the primary segmentation Given an edge set, E, create an edge

graph E’ such that for each edge (x,y) in E, there is a vertex xy in E’

Annotations are associated with regions of primary data by referencing the edge graph vertices Annotations never reference the primary

data directly

Edges in E’ are defined when annotations reference vertices in E’ Vertices may or may not be contiguous

An annotation is associated with vertices in E’ as follows:

1. Create a new vertex, v2. Label it with the FS containing the annotation

content3. Create an edge from v to 0 or more vertices in E’

Zero reference is used in the special case where the annotation applies to information not present in the data

References to 2 or more vertices in E’ by by default concatenate the information covered by the referenced vertices (in order)

can be overridden to specify vertices are to be regarded as an ordered list or “bag”

Edge graph over primary data

|T|h|e| |c|l|o|c|k| |s|t|r|u|c|k| |t|w|e|n|t|y|-|t|w|o| |

Annotations associated with vertices

in the primary data edge graph

type=tokenpos=nnbase=clock

type=tokenpos=detbase=the

type=tokenpos=vbdbase=strike

type=tokenpos=cdbase=twenty+two

As many annotations as desired can reference the same segmentation or be layered over lower-level annotations

SEG2

Primary data

MS1

MS2 NP

Syn2

Co-Ref

Syn1SEG1

MS3 Sem

Annotating Annotations Vertices in an annotation may be

referenced from other annotations1. Create a new vertex, v’2. Label it with the FS containing the annotation

content3. Create an edge from v’ to one or more

vertices associated with an annotation The strategy described above may be

applied recursively, thus creating a DAG whose leaves are the vertices in E’

Annotations associated with token annotations

type=npnumber=sing

type=vptense=past

type=npnumber=sing

1 2 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7

type=tokenpos=nnbase=clock

type=tokenpos=detbase=the

type=tokenpos=vbdbase=strike

type=tokenpos=cdbase=twenty+two

|T|h|e| |c|l|o|c|k| |s|t|r|u|c|k| |t|w|e|n|t|y|-|t|w|o| |

XML Instantiation

<!-- edges over primary data --><edge id="e1" from="0" to="3"/><edge id="e2" from="4" to="9"/><edge id="e3" from="10" to="16"/><edge id="e4" from="17" to="23"/><edge id="e5" from="23" to="24"/><edge id="e6" from="14" to="27"/>

Token Annotation

<node id="t2" edgesTo="e2"> <fs type="token"> <f name="base" value="clock"/> <f name="pos" value="NN"/> </fs>

</node>

Creates a new vertex (node) associated with the FS with a single edge to vertex “e2” in the primary segmentation edge graph

NP Annotation

<node id="np1" edgesTo="t1 t2"> <fs type="NP"> <f name="number" sVal="singular"/> </fs></node>

Creates a new vertex (node) associated with the FS with two outgoing edges to vertices “t1” and “t2” in the token annotation

Question

ISO TC37 SC4 - WG 1Beijing 2006

When referring to annotations, edge targets typically represent components E.g. in the example: “the” and “clock” are

components of “NP” But this is not always the case

Could be e.g. a list of co-referents Others?

Possible solution: let the processor deal with it using the FS type

Note

ISO TC37 SC4 - WG 1Beijing 2006

Edges are never labeled, unlike in many linguistic analyses Preserves simplicity of the graph Relations are DatCats

edgesTo attribute can be empty Can create pseudo-nodes Implies a flat (non-nested) structure in the

dump format

ISO TC37 SC4 - WG 1Beijing 2006

<node type="clone" id="E2" ref="t2"/><node id=“c5” edgesTo=“t5”><f name=“role” sVal=“gen”/></node> <node id=“c6” edgesTo=“t2”><f name=“role” sVal=“head”/></node><node id=“c7” edgesTo=“c5 c6”/><node id=“c1” edgesTo=“t1”><f name=“role” sVal=“head”/></node><node id=“c7” edgesTo=“c7”><f name=“role” sVal=“s”/></node><node id=“c3” edgesTo=“t3”><f name=“role” sVal=“obj”/></node><node id=“c4” edgesTo=“E2”><f name=“role” sVal=“subj”/></node><node id=“D1” edgesTo=“c1 c7 c3 c4 E2”/>

objheads

FLEAHAVEheadgen

subj

DOGMY

[DOG]

Advantages of DAG

ISO TC37 SC4 - WG 1Beijing 2006

Can apply graph algorithms to traverse the graph Breadth-first, depth-first traversal, shortest path, minimum spanning

tree Connectedness, articulation vertices Topological sort Graph coloring, graph partitioning Etc.

What can we do with this? What is all info on path to/from node x What is nearest common ancestor of nodes x and y Find matching sub-graphs Identify connected components Which nodes (phenomena) are most connected, form articulation

vertices, etc. …

Feature Structures

ISO TC37 SC4 - WG 1Beijing 2006

Each edge is labeled with a feature value Can be FS, collection (list, bag, set), atom

Alternation and grouping handled by the FS mechanisms

Need to identify “basic” FS mechanisms 90% of annotations use only these Annotations may (optionally) use only this set

Ease of use No need to implement procedures to handle full power

of FS Need to create a FS library for abbreviation

Implications for Other WGs

ISO TC37 SC4 - WG 1Beijing 2006

Should (conceptually at least) separate referential structure from annotation content E.g. “tlink” in TimeML/SemAF: the link itself is the

edge, “tlink” is the annotation content (?) Need for coordination

Inter-project coordination committee? Need examples!

Today’s Work

ISO TC37 SC4 - WG 1Beijing 2006

Discuss the format in terms of specific annotation types Remember that dump format is in principle never

seen by the user Map user format into and out of dump format

Two topics DAG for referential structure FS for representing annotation content