How to encode compact/compressed headwords

[daliboris]

In Czech printed dictionaries it is not uncommon to encode variants of the lemma in one word, for example

byte(d)lný, i.e. bytedlný and bytelný (residing) (also used as a running head)

(j)herce, jhercě and hercě (player; see also uncompressed headwords jhráč and hráč - could be compressed as (j)hráč)

It would be possible to use @expand attribute of the <orth> element, where space is a lemma separator:

<form type="lemma">
  <orth expand="jhercě hercě">(j)hercě</orth>
</form>

But not in the case of

leleti (sě), leleti and leleti sě (sway and sway itself)

where sě is part of the reflexive variant of the lemma

<form type="lemma">
   <orth expand="leleti leleti sě">leleti (sě)</orth>
</form>

What are othe possible (TEI Lex-0 v. 0.9.3 valid) solutions:

<form type="lemma">
   <orth type="abbreviated">leleti (sě)</orth>
   <orth type="expanded" resp="#boris">leleti</orth>
   <orth type="expanded" resp="#boris">leleti sě</orth>
</form>

<form type="lemma" subtype="abbreviated">
 <orth type="abbreviated">leleti (sě)</orth>
</form>
<form type="lemma" subtype="expanded" resp="#boris">
 <orth type="expanded">leleti</orth>
</form>
<form type="lemma" subtype="expanded" resp="#boris" >
 <orth type="expanded">leleti sě</orth>
</form>

What would you recommend?

In the last example, the compressed variants are also part of word forms (for 1st and 2nd person: -eju, -éš (sě)) and definition (vlnit (se)), so the chosen encoding should be also applicable for <form type="inflected"> and <def> elements.

[ttasovac]

We discussed this during our JF in March (see notes). Subsequently, I came up with some additional thoughts which I also discussed with Laurent on March 16.

But to cut the long story short: I'm proposing that we treat condensed/compacted lemma a là leleti (se) as a labels rather than a lemma because there is no such thing as a naturally occurring word form which contains the brackets. It would, therefore, be wrong to say that "leleti (se)" is itself a lemma.

Instead, I think we should go for:

<form type="lemma">
  <lbl>leleti (sě)</lbl>
  <form type="lemma" resp="#boris">
    <orth value="leleti"/>
  </form>
  <form type="lemma" resp="#boris">
    <orth value="leleti sě"/>
  </form>
</form>

• because word forms leleti and leleti sě do not appear in the print edition, we should use the @value attribute on orth which "gives a value which lacks any realization in the printed source text"
• the drawback here is that regular lemmas (which appear in the print edition) and reconstructed lemmas (which do not appear in the print edition) will live at different paths (//form[@type='lemma']/orth vs. //form[@type='lemma']/orth/@value) but I think we need to accept that because we are reconstructing lemmas that are not in the original, yet we are making it possible to search for individual reconstructed forms this way...
• we can explain in the documentation that indexing all lemmas in a particular dictionary may require alternative expressions such as //form[@type='lemma']/orth/@value|text()

[daliboris]

I tried to apply the <lbl> approach to the entire entry with the existing result:

<entry xml:lang="cs-x-old" type="mainEntry" xml:id="MSS.leleti">
 <lbl>leleti (sě)</lbl>
 <form type="lemma" resp="#boris"><orth value="leleti" /></form>
 <form type="lemma" resp="#boris"><orth value="leleti sě" /></form>
 <metamark function="delimiter">, </metamark>
 <lbl type="short">-eju, -éš (sě)</lbl>
 <form type="inflected" resp="#boris">
  <orth value="leleju" />
  <gramGrp>
   <gram type="person" value="1" />
   <gram type="number" value="singular" />
   <gram type="tense" value="present" />
   <gram type="mood" value="indicative" />
  </gramGrp>
 </form>
 <form type="inflected" resp="#boris">
  <orth value="lelejéš" />
  <gramGrp>
   <gram type="person" value="2" />
   <gram type="number" value="singular" />
   <gram type="tense" value="present" />
   <gram type="mood" value="indicative" />
  </gramGrp>
 </form>
 <form type="inflected">
  <orth value="leleju sě" resp="#boris" />
  <gramGrp>
   <gram type="person" value="1" />
   <gram type="number" value="singular" />
   <gram type="tense" value="present" />
   <gram type="mood" value="indicative" />
  </gramGrp>
 </form>
 <form type="inflected">
  <orth value="lelejéš sě" resp="#boris" />
  <gramGrp>
   <gram type="person" value="2" />
   <gram type="number" value="singular" />
   <gram type="tense" value="present" />
   <gram type="mood" value="indicative" />
  </gramGrp>
 </form>
 <gramGrp>
  <gram type="aspect" expand="nedokonavé" value="imperfective">ned.</gram>
 </gramGrp>
 <sense xml:id="MSS.leleti.sense.1">
  <lbl>vlnit (se), kolébat (se)</lbl>
  <def resp="#boris" value="vlnit, kolébat" />
  <def resp="#boris" value="vlnit se, kolébat se" />
 </sense>
</entry>

The above result raised the following questions:

1. Should we recommend the @type value of the <lbl> element (for example <lbl type="condensed">)? For example, to avoid indexing this element (or to use special treatment when indexing), since the same/similar values can be obtained from @values attributes.
2. Should we apply the identical approach to the other parts of the dictionary entry because there is no such thing as a naturally occurring word form which contains the brackets? See <sense>/<def>.
3. Should the @resp attribute be part of the <form> element, or <orth>? In most cases, grammatical information (such as person, number...) is implicit in the text (it can be inferred from occurring word forms) - I usually don't add the @resp attribute in this case, but using the @resp attribute for the <form> element means that all descendant elements, including <gram>, inherit this attribute and its value.

[laurentromary]

Just a quick note. It occurs to me that <lbl> (lemma, inflections etc.) should indeed be typed so that a publication platform would know what to do with those objects.

[ttasovac]

I missed @laurentromary's comment on typing <lbl> — I agree fully. I will add the type to our example (it's in PR #244, not yet merged anywhere).