question about spanish MOR

[Janet Y Bang]

Hello, 

We are working with the Spanish MOR and noticed a few errors in the MOR line: 

1. bonita oftentimes gets broken down into co:voc|bonita=pretty in cases where the word "bonita" is being used to modify a noun, e.g., utterance is "niña bonita. 
2. In cases were "rato" or "ratito" is used to indicate a period of time, oftentimes the breakdown is n|rato&m-DIM=rat OR n|rato-m=rat. 
   

What would be the most efficient way to fix these errors? We have a lab-internal .cut file where we are adding new words to our internal mor dictionary that were not in the downloaded dictionary (so that we can keep track of differences between our lab lexical items and the downloaded dictionary), but we weren't sure how to override words that were already in the dictionary? Would it be best to fix these in the relevant cut files in the lex folder (i.e., adj.cut, n.cut, respectively)? 

Thank you,

Janet

--

Janet Y. Bang, Ph.D.

Postdoctoral Fellow

Department of Psychology

Stanford University

jb...@stanford.edu

[Brian MacWhinney]

Dear Janet,

    Typically, automatic taggers get about 95% of the words right.  The best for adult written English are up to 97%.  For CHILDES English, we can hit about 97%.  For Spanish, we are probably down around 95%, although sometimes it seems better.  

In the case of ratito, you must be using some much earlier version of SPA MOR.  The current version doesn't have this problem.  The case of bonita is different. It created a few test sentences and it seems that this problem arises when bonita is the last word in the sentences, as in "es una niña bonita".  The problem is that the association of "co" with final position is so strong that it overrides the association of an adjective with the preceding noun.  This can be corrected with a couple of prepost rules that take words that can be either co or adj and force them to be adj when they follow a noun.  Here are the rules:

# es una gata bonita

n|* co:voc|*^adj|* => n|* adj|*

# es una niña bonita

co:voc|*^n|* co:voc|*^adj|* => n|* adj|*

These work for these cases, but (1) you may have other cases I don't know about, and (2) prepost rules can also mess up other things.  So, please give this a try.

--Brian

-- 
You received this message because you are subscribed to the Google Groups "chibolts" group.
To unsubscribe from this group and stop receiving emails from it, send an email to chibolts+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/chibolts/DM5PR02MB3275154B3A345DAA4BD9B5A2D7310%40DM5PR02MB3275.namprd02.prod.outlook.com.

[Janet Bang]

Hi Brian, 

Thank you for your help! Yes we have noted that for Spanish that the taggers are quite good in terms of the break down. In our own review of a couple of transcripts we've also noted about 95% accuracy for part of speech and lemma each and even a higher percentage if you consider accuracy of the lemma alone, so we are pretty happy with how we can automate such a complicated step. 

Our version of the SPA MOR was downloaded in 2018, so we will download the newer version. 

For the prepost rules, I am not familiar with how to incorporate these. How would I include this in the SPA MOR?

Janet

To view this discussion on the web visit https://groups.google.com/d/msgid/chibolts/ECE843B9-F0F4-4D19-920D-BE7D4DD5F425%40andrew.cmu.edu.

[Brian MacWhinney]

Dear Janet,

The use of prepost rules is described in the MOR manual. However, I wouldn’t recommend creating them unless you have a really clear idea about all the ways in which they could overgeneralize.

-- Brian MacWhinney
Teresa Heinz Professor of Cognitive Psychology,
Computational Linguistics,
and Modern Languages, CMU

> To view this discussion on the web visit https://groups.google.com/d/msgid/chibolts/CAC5V4hjo%3DopXfGrbh3r1mPBRhJyubDVkD2nCP7f9a_7wBwmEfw%40mail.gmail.com.

[Janet Bang]

Got it. We will try it out and see if it will work for our transcripts. Thank you!

To view this discussion on the web visit https://groups.google.com/d/msgid/chibolts/8B7FBEEB-CD2D-4FA5-A903-3609DF85E2D3%40andrew.cmu.edu.