[developers] LOGON batch translation profile flags

[Eric Nichols]

Greetings,

It appears that AID is actually what I was looking for. I got confused  
because the first couple of sentence in my fan only had translations  
for the first parse. If my analysis of the information in result's  
flags list is correct, I would like to document it on the Wiki. Could  
someone fill in the gaps for me?

Thanks again,

Eric

On Wed, Jun 3, 2009 at 5:20 PM, Francis Bond <fcbond at gmail.com> wrote:
> G'day,
>
> Isn't (:nanalyses . 5) the number of analyses overall, and RID the
> result ID from the parsing stage, that is the parse ID that you are
> after?  Note that RID may be non-contiguous, as parses that didn't
> transfer are skipped.
>
> 2009/6/3 Eric Nichols <ericnichols79 at gmail.com>:
>> Greetings,
>>
>> I am trying to pull the parse ID out of the result file from a
>> translation profile made with LOGON's batch script.
>> The number of analyses appears to be stored in parses's flags:
>>
>> (:nanalyses . 5) ...
>>
>> However, looking in results, I cannot find any field that appears in
>> the range of the number of analyses.
>> Looking in result's flags, I find the following, which I have
>> commented to my best understanding.
>>
>> ((:WA . 0.22222222)     ;;; word acccuracy based on edit distance.  
>> see
>> $LOGONROOT/ntnu/bleu/wa.pl
>>  (:TORBJOERN . 0.20808958)      ;;; ntnu's bleu score implementation?
>> see $LOGONROOT/ntnu/bleu/bleu.pl
>>  (:BLEU . 0.0)  ;;; bleu score. see $LOGONROOT/nist/mteval-v11b.pl
>>  (:NEVA . 0.0827)       ;;; neva score. see $LOGONROOT/nist/neva.pl
>>  (:RID . 6)             ;;; result-id
>>  (:TID . 2)             ;;; transfer-id?
>>  (:AID . 0)             ;;; should be analysis (parse) id, but looks
>> like it is also result-id?
>>  (:LNF . 0)             ;;; norwegian->foreign GIZA++ probability?
>>  (:LFN . 0)             ;;; foreign->norwegian GIZA++ probability?
>>  (:ASCORE . 6)  ;;; analysis (parse) score?
>>  (:UNKNOWN . 0) ;;; mark if unknown?
>>  (:NMTRS . 13)  ;;; number of MTRs
>>  (:MTRS ZERO_ARG1_REL_NOUN_EF PROGRESSIVE_NOM_EF WA_D_REL_DITCH_EF
>> PRONOUN_Q UDEF-DEF_UDEF_A_Q MARK_OF_REL_NOUN_EF
>> KANKEI_S--CONNECTION_N_OF--EDICT-MTR
>> UNDOU_S_1--EXERCISE_V_1--MOSES-OMTR STD_PRON_N
>> RELATION-NOMINALIZATION_REL-MTR SONO-THAT_Q-MTR
>> NI_P+EXIST_V-HAVE_V-MTR VN-N_JF) ;; list of MTRs
>>  (:TRATIO . 0.72727275) ;;; ???
>>  (:TSCORE . -2550.76)   ;;; transfer score
>>  (:RSCORE . -7.4229403) ;;; realization (translation generation)  
>> score?
>>  (:DISTANCE . 42)       ;;; ???
>>  (:PERPLEXITY . 768.14) ;;; ???
>>  (:LM . 76.6818)        ;;; language model score
>>  (:RANK . 26)   ;;; rank among translations
>>  (:SCORE . -41.521675)  ;;; ???
>>  (:RATIO . 0.6666667)   ;;; ???
>>  )
>>
>> It seems logical that :AID would contain the parse ID, but it's range
>> doesn't match the number of analyses
>> given in parse, and looking at the source in src/tsdb/mt.lisp, it
>> appears to be set to result-id:
>>
>> src/tsdb/mt.lisp:237:        for aid = (get-field :result-id result)
>>
>> Could somebody let me know if the parse ID is stored during batch
>> translation, and if so, how I can get it from the profile?
>>
>> Thanks,
>>
>> Eric
>>
>
>
>
> --
> Francis Bond <http://www2.nict.go.jp/x/x161/en/member/bond/>
> NICT Language Infrastructure Group
>