A formal framework for linguistic annotation

TLDR: A wide variety of existing annotation formats are surveyed and a common conceptual core, the annotation graph, is demonstrated, which provides a formal framework for constructing, maintaining and searching linguistic annotations, while remaining consistent with many alternative data structures and file formats.

About: This article is published in Speech Communication.The article was published on 2001-01-01 and is currently open access. It has received 398 citations till now. The article focuses on the topics: Annotation & File format.

Figures

Figure 13: Possible Structures for a Single Layer

Figure 4: Graph Structure for CHILDES Example (Version 2)

Figure 15: Inter-Arc Linkages for Parallel Transcriptions

Figure 6: Graph Structure for LDC Broadcast Transcript Example

Figure 14: Annotation Graphs Enriched with Inter-Arc Linkages

Figure 12: Gestural Score for the Phrase ’ten pin’

Frequently asked questions

Q1. What are the contributions in "A formal framework for linguistic annotation" ?

This paper focuses instead on the logical structure of linguistic annotations. The authors survey a wide variety of existing annotation formats and demonstrate a common conceptual core, the annotation graph. This provides a formal framework for constructing, maintaining and searching linguistic annotations, while remaining consistent with many alternative data structures and file formats. This technical report is available at ScholarlyCommons: https: //repository. 

Q2. What are the future works mentioned in the paper "A formal framework for linguistic annotation" ?

More complex modes of interaction are also possible, as are connections to other sorts of databases ; the authors regard this as a fruitful area for further research. 

Q3. What are the possibilities of gaps and overlaps?

Gaps might correspond to periods of silence, or to periods in between the salient events, or to periods which have yet to be annotated. 

Q4. What is the tipter architecture for linguistic annotation of text?

The Tipster Architecture for linguistic annotation of text [19] is based on the concept of a fundamental, immutable textual foundation, with all annotations expressed in terms of byte offsets into this text. 

Q5. What is the way to define annotation graphs?

Based on the formal precedent of SGML, the model of how chart-like data structures are actually used in parsing, and the practical precedents of databases like TIMIT, it is tempting to consider adding a sort of grammar over arc labels as part of the formal definition of annotation graphs. 

Q6. What is the common way to break an annotation graph into chunks?

The authors will usually need to break an annotation graph into chunks which can be presented line-by-line (much like interlinear text) in order to fit on a screen or a page. 

Q7. What is the purpose of the definition of anchored annotation graphs?

For the sake of a clean algebraic semantics for the query language, the authors will permit queries and the results of queries to be (sets of) arbitrary annotation graphs. 

Q8. What is the direct way to treat time references?

The most direct way would be to treat Tipster byte offsets exactly as analogous to time references – since the only formal requirement on their time references is that they can be ordered. 

Q9. What is the main problem with existing tools and formats?

While the utility of existing tools, formats and databases is unquestionable, their sheer variety – and the lack of standards able to mediate among them – is becoming a critical problem.