Alignment appears random

[Macho Philipovich]

Hi there,

I installed pialign and the first couple of alignments that I tested appear to be no better than would be a random answer.

I chose an arbitrary sentence from a Canadian statute, namely:

• First Nation means a band, or an Indigenous group that is party to a self-government agreement implemented by an Act of Parliament.
• première nation S’entend soit d’une bande, soit d’un groupe autochtone qui est partie à un accord sur l’autonomie gouvernementale mis en oeuvre par une loi fédérale.

The answer pialign gave me is the following:

• < [ ((( première ||| by ))) [ [ ((( nation ||| an ))) [ ((( S’entend ||| Act ))) { < ((( soit ||| of ))) ((( d’une |||  ))) > } ] ] ((( bande, ||| Parliament. ))) ] ] [ [ [ ((( soit ||| First ))) [ ((( d’un ||| Nation ))) [ [ ((( groupe ||| means ))) < ((( autochtone ||| a ))) < [ [ ((( qui ||| band, ))) [ < ((( est |||  ))) < ((( partie ||| an ))) ((( à ||| or ))) > > < < [ ((( un ||| group ))) { < ((( accord ||| that ))) ((( sur |||  ))) > } ] ((( l’autonomie |||  ))) > ((( gouvernementale ||| Indigenous ))) > ] ] [ < ((( mis ||| party ))) ((( en ||| is ))) > ((( oeuvre ||| to ))) ] ] ((( par ||| a ))) > > ] ((( une ||| self-government ))) ] ] ] ((( loi ||| agreement ))) ] ((( fédérale. ||| implemented ))) ]
  

I tried a second test sentence with pialign, in case this was some kind of outlier, but the result was similar. I don't expect perfection, but this output appeared random.In disappointment, I turned to mgiza++, but its out-of-the-box response to the same input was even more incomprehensible, namely:

• # Sentence pair (1) source length 24 target length 36 alignment score : 4.33196e-97
  première nation s ’ entend soit d ’ une bande , soit d ’ un groupe autochtone qui est partie à un accord sur l ’ autonomie gouvernementale mis en oeuvre par une loi fédérale .
• NULL ({ 2 3 5 10 }) first ({ 29 30 31 32 36 }) nation ({ }) means ({ }) a ({ 33 34 35 }) band ({ }) , ({ 4 }) or ({ 6 }) an ({ }) indigenous ({ }) group ({ }) that ({ }) is ({ 11 12 13 15 16 17 18 19 22 }) party ({ 20 21 23 24 25 27 28 }) to ({ }) a ({ 9 }) self-government ({ }) agreement ({ }) implemented ({ }) by ({ }) an ({ 1 }) act ({ }) of ({ 7 }) parliament ({ }) . ({ 8 14 26 })

Is there anything I can do to get helpful results? I looked at other word alignment tools, namely eflomal, the Berkeley aligner, and tdx-nlp but I wasn't able to get them to either compile or run. The Berkeley aligner, in particular, required a tree format version of the input, and its docs say a script to generate this would be released later, so that stumped me.

Any help in getting reasonable word alignment working would be greatly appreciated.

Many thanks,

Macho

[Leather Dog Muksihs]

How large is your bi-lingual corpus? [line count]

[Macho Philipovich]

Thanks for the response.

I'm really sorry to have caused the trouble, but at the time I'd written this I fundamentally misunderstood what sentence aligners do, and my entire corpus was the text I included in the email. Have been using bigger a corpus since then and getting more reasonable results.

Best,

Macho

--
You received this message because you are subscribed to the Google Groups "pialign-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pialign-user...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/pialign-users/5fcb6922-ac17-4a92-a865-394868669fe4%40googlegroups.com.