Magento CommerceMagento Commerce

computerized language sample analysis


Bound morphemes combined, error in total in SALT

Hi!

I found that in "house/s/z", SALT total morphemes is 3 instead of 2. Because the program adds 1 per "/s", 1 per "/z" and 1 per "/s/z". This overinflates MLU (in morphemes). The same happens in Spanish when we try to add "/a" morpheme to mark feminine, before adding "/s".

Would it be able to solve this problem? I don't know if there any reason because is in this way. I guess a rule which excludes any combination of more than one morpheme in the total account would be very useful.

Thanks a lot in advance,

Originally posted by Magdalena Capo

10 thoughts on “Bound morphemes combined, error in total in SALT”

  • Salt Support

    The 3rd morpheme is the word root. There are 2 bound morphemes but 3 morphemes.

    house = 1 morpheme
    house/s = 2 morphemes
    house/s/z = 3 morphemes

    Reply
  • Salt Support

    Thank you a lot for your response. But if it is in this way, why this rule is not valid for all the words?

    I mean, if house/s/z is (as you replied above)

    house= 1 morpheme

    house/s= 2 morphemes

    house/s/z= 3 morphemes

    The logic rule would be

    house/s

    house= 1 morpheme

    house/s= 2 morphemes

    But in SALT is

    house/s= 1 morpheme. (/s)

    And this has an impact in MLU in morphemes calculation. Why is that?

    Thank you a lot,

    Originally posted by Magdalena Capo

    Reply
  • Salt Support

    I may be missing your point. If you create a 1-utterance sample with 1 word, e.g.,

    $ child
    C house/s.

    You will see in the Word and Morpheme Summary that the MLU in morphemes = 2. And if you look at the Utterance by Utterance Length table, you will see that there is 1 utterance with 2 morphemes.

    Reply
  • Salt Support

    Thank you. You are right. But if you check Bound Morpheme tables there's the difference I told you before, and I don't know if my SALT software is doing ok or not, but the thing is the final MLU is ok, so is the same you have reported.

    C house/s.

    Bound Morpheme Table
    Total
    /s
    house/s 1

    C house/s/z

    Bound Morpheme Table
    Total
    /s
    /s/z
    /z
    house/s/z 3

    Thank you a lot again for your clear explanations.

    Originally posted by Magdalena Capo

    Reply
  • Salt Support

    Consider the same 1-word transcript as before, i.e., C house/s/z.

    If you generate the Bound Morpheme Table, NOT expanded by words or by words and codes, you will see:

    /s - 1 occurence
    /z - 1 occurence

    If you expand it by words, you will see:

    /s - 1 occurence (house/s/z)
    /s/z - 1 occurence (house/s/z)
    /z - 1 occurence (house/s/z)

    When expanded by words, it may give the appearance that there are 3 different bound morphemes since /s/z is shown combined and also shown separated into /s and /z. Perhaps that is the confusion.

    Reply
  • Salt Support

    Great! This means that these combinations like "/s/z" are not counting as bound morphemes but just as ocurrences, and this is why MLU is not affected by!! Many many thanks!

    I guess it will be the same for Spanish. Because we want to add more bound morphemes as diminutives(see /it below) and feminine morphemes(/a) as in


    perritas | perro/it/a/s (dog/diminutive/femenine/plural)

    or

    perra | perro/a (dog in femenine)

    But specyfying word roots(types) is so much time consuming in Spanish. Is there any other way to obtain a correct Type-Token-Ratio through specifying bound morphemes with word roots?

    The thing is that in Spanish language "perro" (dog) is considered the Type for perra (dog in femenine), perritas (doggies in Femenine), and all the forms derived from "perro". There is a lot percentatge of words which are working in this way. So, perra (meaning dog in femenine), is not a valid type, is a token.

    Best Regards and many thanks again!

    PS: when we will finish the project, we will be delighted to share all the data results with other researchers. We are going to obtain data from clinical and control groups of monolinguals in Spanish from narratives and conversations.

    Originally posted by Magdalena Capo

    Reply
  • Salt Support

    For the Spanish database samples in SALT, we did not mark diminutives and gender as bound morphemes. The only bound morpheme we marked was the plural.

    You could create your own Root Identification File (RIF) to assist with marking root forms as you want them marked. If you could provide an Excel spreadsheet with 2 columns, the produced form and the identified form for your most common words, we would work with you to create a RIF for your use.

    An example of the Excel info:
    Column A: peritas
    Column B: perro/it/a/s

    Column A: perra
    Column B: perro/a

    If you are interested in pursuing this, email the spreadsheet to Ann at [email protected]

    Reply
  • Salt Support

    Thanks a lot!

    Yes! I'm interested in doing so, but in the meantime, what do we do with types, morphemes and woord roots? do we let unmarked? i.e. C perritas.

    We think the list you suggest it would be feasible in the end, but some problems arise:

    1) For narrative context is ok but a bit difficult beforehand for conversation task due to this is topic free and not guided. We think we would have an approximate list in the halfway of transcription process.

    2) Linguistic constraints: for verbs, determiners (and some pronouns) is easier than for nouns and adjectives because these last ones are opened cathegories and the most common combinations with morphemes are a lot if we want to cover Spanish language. But we would do it based on our data.

    I will email to Ann in a fortnight.

    Originally posted by Magdalena Capo

    Reply
  • Salt Support

    Apart from legibility, which decreases, is there any reason for not to be able to use column B (real word root and morphemes) as direct transcription of types? i.e. Child says perritas and instead of transcribing

    C perritas| perro/it/a/s.

    why not

    C perro/it/a/s.

    We have just checked for word roots, bound morphemes, TTR and MLU and it seems that the results are the same.

    Originally posted by Magdalena Capo

    Reply
  • Salt Support

    Good point. For regular forms, there would be no need to use the vertical bar. You are correct in that they both yield the same result.

    Reply
Leave a Reply
© Salt Software LLC. All Rights Reserved.