Spatio-Temporal Parsing in Spatial Hypermedia

Abstract

Spatial Hypertext represents associations between chunks of information by spatial or visual attributes (such as proximity, color, shape etc.). This allows expressing information structures implicitly and in an intuitive way. However, automatic recognition of such informal, implicitly encoded structures by a machine (a so-called spatial parser) is still a challenge. One reason is, that conventional (non-adaptive) parsers are conceptually restricted by their underlying source of information (i. e., the spatial hypertext). Due to this limitation there are several types of structures that cannot be recognized properly. This inevitably limits both quality of parser output and parser performance. We claim that considering temporal aspects in addition to spatial and visual properties in spatial parser design will lead to significant increase in parsing accuracy, detection of richer structures and thus higher parser performance.For the purpose of providing evidence, parsers for recognizing spatial, visual and temporal object relations have been implemented and tested in a series of user surveys. It turned out, that in none of the test cases pure spatial or visual parser could outperform the spatio-temporal parser. Instead, the spatiotemporal parser was able to compensate limitations of conventional parsers. Furthermore, differences in parsing accuracy were successfully tested for statistical significance. The results indicate a non-trivial effect that is recognizable by humans. We have shown that the addition of a temporal parser shifts machine detected structures significantly closer to what knowledge workers intend to express.iii ResuméSpatial Hypertext repraesenterer associationer mellem informationsdele fra spatiale eller visuelle attributter (såsom naerhed, farve, form osv.) Dette gør det muligt at udtrykke informationsstrukturer implicit og på en intuitiv måde. Men automatisk genkendelse af sådanne uformelle, implicit kodede strukturer (ved anvendelse af en såkaldt spatial parser) er stadig en udfordring. En af grundene er, at konventionelle (ikke-adaptive) parsere begrebsmaessigt er begraenset af deres underliggende informationskilde (dvs. den spatiale hypertekst). På grund af denne begraensning, er der flere typer af konstruktioner, der ikke kan genkendes korrekt. Dette begraenser uundgåeligt både parser output kvaliteten og parser ydeevnen. Vi haevder, at inddragelse af temporale aspekter i tillaeg til spatiale og visuelle egenskaber i spatial parser design vil føre til betydelig øgning af parsing nøjagtighed samt detektering af rigere strukturer og dermed højere parser ydeevne.Med henblik på at vise dette, er parsere for genkendelse spatiale, visuelle og temporale objekt relationer implementeret og testet i en raekke brugerundersøgelser. Det viste sig, at i ingen af testcasene var en ren rumlig eller visuel parser mere effektiv end den spatio-temporale parser. I stedet var den spatiotemporale parser i stand til at kompensere for de konventionelle parseres begraensninger. Endvidere blev forskelle i parsing nøja...

Mother

“…Such implicit structures are supported by a spatial structure service in Asgard. It includes several multiple, highly specialized parsers (spatial, visual, temporal) that are capable of analyzing spatial structures similar to how humans would interpret them [35].…”

Hypertext as Method

“…This is not what many would expect from a computer, as this results in fuzzy connections, ambiguities, or different interpretations. Such implicit structures are more difficult to handle by machines [46], which can result in people not expecting to work with such structures on computers.…”

“…Spatial hypertext systems often consider multiple parses of a particular space [cf. 46]. In hypertext literary work, there are discussions of traversaldependent links [cf.…”

“…The most distinguished feature in Mother is its spatial parsers that are capable of computing associations between nodes placed on a 2D canvas. This is based on their visual cues, spatial arrangements, or distance [46]. Figure 2 shows Mother's architecture.…”

“…Cubbles are related to each other. The spatial structure service in Mother's software layer Asgard uses spatio-temporal parsers [46] that analyze relations between components that the user has dropped onto the BDE. From that, the system acquires related Cubbles from the knowledge base and presents them to the user.…”