|Number||Description||Scale||Suggested By||Suggested For|
|1||Implement full SADL. This is important for the work at UMASS as well as improving the expressiveness of SHAKEN's KR language. It will require enhancements to KM as well as the core library.||2 m-m||Bruce||team wide|
|2||Fill out the representations of the core components of the library. They are currently skeletal, but we know how to improve them significantly, especially using constructs from SADL and default rules.||6 m-m||Bruce||UT|
|3||Implement default rules in KM.||Bruce||Pete|
|4||Improve the text generator in SHAKEN||medium||Bruce|
|6||The explanation screen (within the Q/A facility) needs work. Here's a good test case: access Ameoba's RNATransciptionGeneral KB. Ask the question: what role does the Polymerase play in the process? (You can select the Polymerase by expanding the Recognize subevent of RNATranscriptionGeneral).||high||Bruce|
The 'auto-expansion of selected nodes' feature is not working
properly, perhaps it's not working at all. Here's a test case: bring
up ATP in a CMAP. Fully expand the root node (ATP). Notice that
Donor is added as leaf node in the graph.
For some slot-fillers, SHAKEN should elaborate the description to an extra level of detail, without the SME asking for it. Here's an initial list of slot-fillers to elaborate.
|14||In the Q/A facility, the pop-up menus show two slots (edge-coordinates and edit-status) that should be hidden from the user.||low||Dan|
|16||KM should use the rule that a slot's fluent-status is the same as the fluent-status of its inverse.||recommendation||Bruce||Pete Clark|
|20||Currently SHAKEN can't handle Property's and Value's. A partial solution might be to enable SHAKEN to display this information, but not enable SME's to enter new information of this type. I have an email thread on this topic, which includes a description of Pete's backend support for Property's and Value's.||moderate||Bruce||Vinay|
In simulating the scenario Car-Accident (which should be available to
you via SHAKEN), a check of the preconditions for Collide revealed no
problems (i.e. all the preconditions succeeded). However, immediately
after KANAL said that the preconditions succeeded, there were a couple
of error messages stating that some preconditions (which supposedly
succeeded) had failed. The output from KANAL looks something like the
Step: Collide Checking preconditions 1.The location of the object of the Collide must be null. The location of the Car must be null. 2.the object of the Collide mustn't be touching the Car mustn't be touching 3.the Light-Pole mustn't be touching -> The three conditions succeeded Checking preconditions 1.the Light-Pole mustn't be touching -> The one condition failed Click here to find help for fixing this error. Checking preconditions 1.the object of the Collide mustn't be touching the Car mustn't be touching -> The one condition failed Click here to find help for fixing this error.This does not seem to be a KB problem because there were no problems when I tested it in KM.
Ken Barker adds the following insight:
|high||Peter||ISI - Jim?|
|26||When specifying expected conditions during the testing of knowledge, Shaken has no facility to allow the user to specify relationships that no longer exist. For example, if I tested a Remove of an engine from a car, I have no way of testing that the engine is no longer part of the car after the action is executed.||high||Peter||ISI|
Yesterday I created two concepts, Pete-2 and Pete-3. There's been
alot of wierdness surrounding these components (which spawned report
24 above), but I'm sending this report so that SRI can confirm whether
there's a bug or SHAKEN or the components are just trashed.
So, here's the report:
Load Pete-3. Attempt to create a relation between Pete-3 and the Recognize subevent. The only relation classes that come up are compliment and subevent. I was unable to specify a 'first subevent' that the tester wants.
b) Here's another report. It's similar to the previous one in that it, too, might be the result of faulty kb-saving, or it might point to a different bug in SHAKEN:
Click on Ask Questions Open Pete-RNA-Chain-Initiation Ask "What is the relation between a Pete-RNA-Chain-Initiation and a Pete-RNA-Transcription?" (Last question in the list) P-R-C-I is a subevent of P-R-T, but KANAL says "No relation found".
| a) closed
b) failed verification (18 June 2001, v 1.14) use P-R-C-I-2 and P-R-T-2
|35||I went to the 'My Concepts' window, and clicked on one of my concepts. That brought up an Inspect Window. The 'go-back' button in the Inspect window seems to have no affect.||reported fixed||Bruce||Sunil|
|36||'prevents' is a slot from Event to Event. It doesn't show up in the list of slots that might be used to connect two events in CMAP. Perhaps this is a KB problem, but I can't see anything wrong with the slot definition.||medium||Bruce||Sunil|
We encountered an issue when using KANAL today. In testing scenarios
that contained branches in the subevent path like the following:
Car-Collision |the output was completely uninituitive. For example, the output for the above scenario looked something like:
Summary of alternative paths simulated Move-To -> Make-Contact -------------------------------------- Step: Move-To Checking preconditions ... Step: Make-Contact Checking preconditions ... Step: Move-To Checking preconditions ... Step: Move-To Checking preconditions ...At first glance, it appeared that KANAL executed a Move-To followed by a Make-Contact followed by two more Move-To. It was only after looking at the KANAL log did it become clear what KANAL was doing (simulating all the possible, legal event paths). The output becomes even more confusing for scenarios that have branches at multiple points in the event path! Thus, we were wondering is it possible to either:
1) Display the output in a manner that will give the user a more intuitive sense of what's going on (e.g. listing a summary of ALL the paths simulated, something like what's currently in the KANAL log). 2) Hide this part of the output for the time being.
I was running a scenario in KANAL today called P-Car-Collision.
When KANAL finished simulating the scenario it outputed the
Steps not simulated due to errors: Move-To , Move-Tounder the section "Other results from the analysis". I went and got the KM code from the KANAL log (which I've sent it in an email message to Yolanda and Jim) and tested it in KM, and there were no errors. The question is: why did KANAL report an error and where did it come from?
When specifying expected conditions during the testing of knowledge,
Shaken has no facility to allow the user to specify relationships that
no longer exist. For example, if I tested a Remove of an engine from a
car, I have no way of testing that the engine is no longer part of the
car after the action is executed.
Is it possible to extend KANAL to allow such effects to be specified?
Just to keep a record of this exchange so we can return to it when
time permits... A SME wrote:
>This SME currently has the two bases (T and U) in one concept map joined at >their superclass. He has found that he can only ask a comparison question >if he creates two separate CMaps (not one joined at the superclass) for >each of the concepts. > >It seems like there ought to be a better way to do this. If so, please >share it.
In the system that is running today, it is not possible to ask the question the way the SME desires. This is a reasonable thing to want to do and could be supported with a little bit of work though. So, we can put it on the list of planned extensions.
PROBLEM: some explanations don't have KB rules explaining the results.
EXAMPLE: Q/A->Ask about properties of DNA-Strand->What is DNA-Strand
notice that the D-TMP part of DNA-Strand does not have explanations.
PROBLEM: the meaning of Structural Function is not clear. EXAMPLE: Description(or Q/A) on Basic-Amino-Acid, notice there's a Structural function entry. It's not clear what structural function refers to.
PROBLEM: the order of subevents in the explanation for a process is independent of the execution order EXAMPLE: Create a process with 3 subevents, and check its description(explanation), the description lists the 3 subevents in the reverse order of creation, not execution/simulation order.
|47||QA cannot answer questions involving the inverse of a slot, although it can answer questions involving the slot. I have reproduced this error both by asking first about the inverse slot, and by asking about the slot and then asking again about the inverse.||low||Dan||QA|
I agree with Pete's suggestion of how to code Transcription, then
RNA, in two steps. This is what Art suggested, too.
However, I think we have some work to do in SHAKEN's Q/A system to properly use this 2-part information. How would we answer these questions:
1) 'describe the process that produces mRNA (say).' Will SHAKEN climb the 'RNA hierarchy' from mRNA to RNA, find Transcription, then describe 'Trascription that produces mRNA'? 2) questions that involve simulating mRNA. Again, will SHAKEN put the two pieces of information together to answer the question?We might need to enrichen our question/answering methods.
This question from the sme called Iflu:
I was trying to answer the question ECB-7.1.1-18 What are the four bases/nucelotides of RNA?
I have found that unless there is a direct relationship between RNA and the 4 bases the system cannot answer the question. I tried putting the bases under nucleotide (hierarchially which is RNA/nucleotide/4 bases) but when I put the same question to the system, it could not answer it. See the attached document.
Celula was trying to ask a question about DNA-melting in her RNA-TranscriptionV2 CMap. At first she was not getting an answer anything like what she expected. I then suggested she ask the question in the context of the RNA-Transcription she had defined. I believe that approach did produce some answers.
Celula writes: I tried to ask the question "What was happen during DNA melting?" (and DNA-melting is defined in the KB) and Shaken answered "something DNA-melting a DNA-Double-Strand, but no more details are known". To me this answer is strange due to the fact that the concept DNA-melting is defined in the KB.
A SME (Iflu) was wondering if it would be possible to have two
directories for work--one for concepts the SME feels good about and
would like to keep; the other for on-going work.
Also, there have been numerous requests from sme's to be able to delete concepts they've created.
This SME, Vaccinia, is struggling with the use of contexts--he has defined
translation and transcription and they are in two separate scenarios. He
know wants to ask a question which involves knowledge of both
scenarios. How should he proceed?
Here's Vaccinia's description of the problem:
I wanted to ask a question to SHAKEN: `Is RNA polymerase a instrument of any event?'
I got the answer "NO"
Then I selected a Scenario and inside the Scenario I asked the same question .... SHAKEN came with a answer "YES"
What I understand by this is the events can be compared from within a scenario, about the participants but we cannot know their function outside the scenario. If we want to compare the participants roles/functions in different scenario's we cannot achieve the same.
One way of overcoming the problem is defining both the scenario's in a single bigger scenario then each of the participants can see the other entity/event for comparion, but that may not be feasible for all steps.
ERROR: Show similarities does not work.
PROBLEM: SHAKEN does not return an answer to the question: "Show similarities between [Copy] and [DNA-Replication]"
This bug was reported also by an SME.