“Tagging Time in PROLOG: from quick and dirty to TEI ”

Jan Christoph Meister University of Hamburg jan-c-meister@uni-hamburg.de

This paper presents work in progress from a current research project at Hamburg University that employs Humanities Computing methodology for developing and testing a new theory and model of “narrative time”. Our premise is that narrative time should be defined in functional and not in essential or categorical terms: time is not an objective phenomenon, but a cognitive construct and can thus best be modeled in terms of a 'temporality effect'. This effect -- that is, the impression of temporal order in narrative, both on the level of fictional reality and narrative discourse – is to be explained and analyzed in terms of the distribution of empirical 'notions' (representations of objects) and 'temporal operators' throughout a representational medium, in our case: a narrative text. Humanities Computing methodology plays a central role with regard to both the description (markup) and the subsequent combinatory analysis of relevant textual elements. However, adhering to a TEI compliant tagging approach proves unacceptably complicated. The paper therefore argues for a quick and dirty approach to time tagging based on feature structure tags that are defined in the form of PROLOG clauses.

THEORY

To date most theories of narrative—in particular those focusing on the domain of literary narratives—conceptualize of ‘time’ in terms of a dichotomy of narrated time vs. time of narration or, as Günter Müller’s classic formulation goes, of Erzählzeit vs. erzählte Zeit. This is essentially an ontological distinction that attempts to set apart two ‘worlds’, each of with is seen to have its ‘own time’. However, this distinction immediately becomes problematic when dealing with non-fictional representations of events which, irrespective of chronological proximity, are by definition situated on a singular objective time line. Our approach is therefore based not on the traditional narratological concept, but rather on the unitary model of time originally proposed by McTavern who distinguished between two perspectives onto time: namely, that of events in an objective before-after relationship (the so-called B-series of time), and that of events as occurring in the subjective cognitive order of future-present-past (A-series).

COMPUTER-BASED IMPLEMENTATION

As far as tools are concerned, the implementation of this theoretical model in a Humanities Computing orientated project has necessitated the development of two programs:

TempusMarker -- a software tool providing automatic and semi-automatic markup routines for the tagging of temporal expressions in natural language texts. A prototype of TempusMarker has already been programmed.°
TempusParser -- an analytical tool that generates a version (or versions, as the case may be) of the base text in which all the sequences that form a complex narrative discourse are organized in strict chronological order. This (re)construction is the result of an algorithm driven process of analysis and recombination of textual segments during which the ‘time stamp’ of each segment as indicated by the temporal tags is interpreted.°

The computational implementation of McTavern’s model for the purpose of concrete analyses of A-series governed textual discourse and its eventual reconstruction in terms of B-series ordered event sequences offers an interesting example for the difficulties faced by the computing Humanist who tries to tackle even modestly ‘intelligent’ hermeneutic problems. In the project to be presented here an added problem stems from its empirical orientation: rather than having experts (Literary Scholars) tag the texts we are using student groups in order to simulate as closely as possible the ‘naïve’ reader’s processing habits. A demonstration of the actual tagging process, including the use of the TempusMarker prototype, will form part of the presentation. The computational implementation of McTavern’s model for the purpose of concrete analyses of A-series governed textual discourse and its eventual reconstruction in terms of B-series ordered event sequences offers an interesting example for the difficulties faced by the computing Humanist who tries to tackle even modestly ‘intelligent’ hermeneutic problems. In the project to be presented here an added problem stems from its empirical orientation: rather than having experts (Literary Scholars) tag the texts we are using student groups in order to simulate as closely as possible the ‘naïve’ reader’s processing habits. A demonstration of the actual tagging process, including the use of the TempusMarker prototype, will form part of the presentation. Whereas the temporal value of the respective natural language expressions -- be they denotative or deictic -- is comparatively easy to establish either contextually, or from a dictionary, their explication in the form of standardized TEI markup (core tag set + additional tag sets for dates and time, ref. chapter 20.4. of TEI guidelines) has proven rather unwieldy and often contra-intuitive to our readers. Defining more readily comprehensible feature structure tags would seem to be the alternative of choice; however, this raises the methodological question of how to design a sufficiently fine-grained feature structure that does not automatically become completely idiosyncratic to the particular research problem at hand, thus inadvertently restricting the uses of the tagged corpus at a later stage. Against this background the paper advocates and will demonstrate a calculated quick and dirty approach to designing temporal feature structure tags. In particular, it will be shown how the PROLOG predicate structure can facilitate rapid prototyping of feature structure tags.

CONCLUSION

Expressing relatively complex hermeneutical problems and models in terms of Humanities Computing methodology and standards should be conceived of as a process of translation, rather than one of mere re-presentation. From a hermeneutical point of view semantic tags are not just descriptors, but rather predicates of a prepositional clause in which the tagged string itself is one argument, and its feature values the subsequent arguments. Capturing experimental temporal feature structure tags in the form of PROLOG predicates therefore holds two advantages: first, it offers a more intuitive approach to semantic tagging. Second, it facilitates automatic conversion of feature structures into composite TEI tags at a later stage, thus turning the quick and dirty into the beautifully intricate -- and fast at that.

REFERENCES

Christopher Habel Frank Schilder. “From Temporal Expressions to Temporal Information: Semantic Tagging of News Messages.” . : ,

Günter Dammann Jan Christoph Meister. “The temporality effect: Design and computer-based application of a constituent model of narrative temporal order.” . : ,

John Ellis McTaggart. “The Unreality of Time.” Mind. 1908. 17: 457-474.

Günther Müller. “Erzählzeit und erzählte Zeit.” Festschrift für Paul Kluckhohn und Hermann Schneider. Tübingen: , 1948. 195-212.

TEI Consortium. TEI P4 - Guidelines for Electronic Text Encoding and Interchange. Ed. C. M. Sperberg-McQueen Lou Burnard. : , 2001.