Incremental Natural Language Description of Dynamic Imagery
read more
Citations
Plan-based integration of natural language and graphics generation
Plan-based integration of natural language and graphics generation
Constructing qualitative event models automatically from video input
VIsual TRAnslator: Linking perceptions and natural language descriptions
Three RoboCup Simulation League Commentator Systems
References
Towards a General Theory of Action and Time.
Towards a general theory of action and time
An Incremental Procedural Grammar for Sentence Formulation
Building, registrating, and fusing noisy visual maps
Related Papers (5)
Frequently Asked Questions (15)
Q2. what is the goal of this work?
Their approach emphasizes concurrent image sequence evaluation and natural language processing, an important prerequisite for real-time performance, which is the long-term goal of this work.
Q3. What is the meaning of course diagrams?
Using course diagrams guarantees that primitive motion concepts as well as complex activities can be defined in an uniform and declarative way.
Q4. What is the predicate used to express the continuation of an event?
In order to model durative events like `move', a further predicate called succeed was introduced to express the continuation of an event.
Q5. What is the purpose of the course diagram?
In addition to a specification of roles denoting participating objects, which must be members of specified object classes, an event model includes a course diagram, used to model the prototypical progression of an event.
Q6. Why does the system have to decide which events to verbalize?
Because of the strong temporal restrictions the system cannot talk about all recognized events, thus it has to decide which events should be verbalized in order to enable the listener to follow the scene.
Q7. What is the process of transforming symbolic event descriptions into natural language utterances?
In the process of transforming symbolic event descriptions into natural language utterances, first a verb is selected by accessing the concept lexicon, and the case-rolesassociated with the verb are instantiated.
Q8. What are the factors that determine the relevance of an event?
The relevance of an event depends on factors like: (i) salience, which is determined by the frequency of occurrence and the complexity of the generic event model, (ii) topicality, and (iii) current state, i.e., events with state succeed or stop are preferred.
Q9. What is the meaning of a course diagram?
Each recognition cycle starts at the lowest level of the event hierarchy: first, the traversal of course diagrams corresponding to basic events is attempted; later, more complex event instances can look at those lower levels to verify the existence of their necessary subevents.
Q10. What is the function of the language generation component?
The language generation component selects relevant propositions from this buffer, orders them and finally transforms the non-verbal information into an ordered sequence of either written or spoken German words.
Q11. What are the main areas of research within AI?
Image understanding and natural language processing are two major areas of research within AI that have generally been studied independently of one another.
Q12. What is the current use of the trajectories?
The as yet partial trajectories delivered by Actions are currently used to synthesizeinteractively a realistic GSD, with object candidates assigned to previously known players and the ball.
Q13. What is the definition of a course diagram?
The recognition of an occurrence can be thought of as traversing the course diagram, where the edge types (:trigger, :proceed, etc.) are used for the definition of their basic event predicates (see Section 3.3).
Q14. how do you use natural language to analyze raw images?
The authors have shown that the various processing steps from raw images to natural language utterances, i.e., picture domain analysis of the image sequence, event recognition, and natural language generation, must be carried out on an incremental basis.
Q15. What is the effect of the action system?
Since the first results described in Schirra et al. [1987], more than 3000 frames (120 seconds) of image sequences recorded from a major traffic intersection in Karlsruhe have been evaluated by the Actions system.