Golfred Phase 1: First in vitro path

Prototype of a Golem textual path
- Due date: August 26
- Members: Iván, Jorge
Printings of the text messages from the prototype path
- Due date: August 28
- Members: Jorge
Photographs of the real texts of the prototype path
- Due date: August 28
- Members: Iván
Proof of concept with ROS.read_text library
- Due date: September 3
- Members: Iván, Jorge
RDF generation with FRED web service
- Due date: September 3
- Members: Iván, Jorge
Text generation from the RDF set following the pattern I WAS IN PLACE X WHERE I READ Y, WHICH MEANS Z
- Members: Iván, Jorge, Claire
- Due date: Setember 30

Preprocesamiento de las cadenas de caracteres buscando entidades nombradas y terminos antes de enviar a Tipalo

Agregar German Open en la lamina del robot
Reconocimiebto de escena posible (informacion exterior al texto)
En la fase de ROS se podria generar un texto rudimentario a partir del mapa de Golem
Antes de entrar el la fase de OCR, hay que hacer una descripcion en primitivas espaciales del mapa
- Entre a
- Sali de
- Di vuelta a la derecha
- Avance
Fred y Tipalo van a llenar los vacios generados en el
- Vi X
- Que hace referencia a Y
Luis propone generar un modelo de dialogo espacial para que Golem describa sus movimientos y sus despazamientos.
Explotar los registros historicos de Golem (historia de la tarea)
Modelo de dialogo muy sencillo (“me movi a”)
Dos fuentes de informacion para Quelo: informacion del espacio (Golem) e informacion del mundo (Tipalo/FRED)
Y si Golem buscara en el directorio del IIMAS?

Named entities with Tagme after OCR
Alignment between the RDF and the sentence
Trim the RDF to find the usegraph useful for generation
The baseline would be the text generation without Quelo
Legalo: RDF summarisation (for RDF generation?)
How to produce the best FRED subgraph for Quelo?
- Motif graph query of FRED
- How much knowledge does the robot needs to know?
Claire proposes a trajectory where we assume that the OCR is perfect and we just generate text from clear messages.
Randomize the trajectory (Luis). The robot could follow people and then read.
Open questions:
- How to improve the OCR
- How to choose the right definition for a text element
- How to generate the right RDF for generation
- What is the best text possible for a fixed path?
- Could be generate text for random path?
What is our real use case?
- Museum exhibit
Afternoon questions

Use case
# As it is
# Simplified, no OCR trajectory
# Random trajectories
# Museum
# Handicapped
# Give narrative cohesion to the generated text around a scientific subject and focus on factual descriptions.
Planning
Mexico travel

Proposed task planning

Talking points
Bureaucratic problems with the project
Iván's web interface as an invitro path
Aldo-Geni-FRED interface
Iván talks about Iván's web interface, which
Aldo taks about scripts for generating Geni code
Before going technical, we discuss about Aldo's seminar at IIMAS.
Aldo shows FRED's service web interface, we talk about Fredlib python interface to query FRED
Aldo shows Json output of FRED
Aldo shows Perl scripts to translate RDF's triples to Geni
Aldo talks about a new FRED gold standard of 150 sentences which verifies that in case of changes and new deployments, FRED's core functions will keep working.
Aldo shows quantifications problems of FRED, and then we go back to the FRED-Geni bridge
We include Emily (Claie's postdoc) in the loop in order to have Geni system.
It would be nice to have students (Mexican, Italian, French) helping us tu put graph gold standards (graph alignment, sentence-graph alignments).
We would like to submit something to IJCAI 2016 workshop on service robots (deadline: April 2016).