Issues with tier hierarchy and tier type

I am preparing a somewhat complex hierarchy for coding multiple, embedded factors in a recorded interview. I have copied it below, with the proposed tier type in brackets next to each tier.

I am new to ELAN, but this seems like an efficient hierarchy for isolating factors via structured searches to produce statistics after coding. All tier types aside from “Speech” and “Language mixture text” would have their own CVs, as there are defined options for each factor.

When I try to assign a child tier to a parent with a different tier type, my options for child tier type disappear. To be specific: the “Speech” tier has its own type “SPEECH”, but no CV because it is transcribed text. Some annotations on this tier can be assigned to a “Language ideology type”, the first child tier, which has its own tier type “LI_TYPE” with a 5-item controlled vocabulary.

However, I can only apply the tier type “LI_TYPE” to the corresponding tier if it is not the child of “Speech”. Does this mean parent and child tiers must have the same tier type? Since CVs can only be assigned via tier type, this seems wrong. I have read through the manual and watched various instructional videos, but I can’t seem to figure this one out.

  • Speech [SPEECH]
    • Language ideology type [LI_TYPE]
  • Language mixture text [LM_TXT]
    • Language mixture type [LM_TYPE]
      • Insertion type [INS_TYPE]
        • Insertion establishment [ESTABLISHMENT]
          • Insertion integration [INTEGRATION]
            • Complexity of determiner phrase [COMPLEXITY]
              • Gender of determiner phrase [DP_GEN]
                • Gender alignment of determiner phrase [DP_GENA]
            • Verb phrase type [VP_TYPE]
            • Complexity of adjective phrase [COMPLEXITY]
            • Complexity of discourse marker [COMPLEXITY]
            • Complexity of bare noun [COMPLEXITY]
              • Number of bare noun [BN_NUM]
      • Calque type [CAL_LS_TYPE]
        • Calque establishment [ESTABLISHMENT]
      • Loanshift type [CAL_LS_TYPE]
        • Loanshift establishment [ESTABLISHMENT]
      • Code switch type [CS_TYPE]
        • Intra-sentential CS type [CS_IA_TYPE]
          • Intra-sentential CS felicity [CS_IA_FEL]
        • Dialogic CS type [CS_DIA_TYPE]


In your example of the “Speech” and “Language ideology type” tiers, they should not have the same type. I guess the point is in the Stereotype of the type you wish to use for the dependent tier. Any top-level tier should have a tier type with Stereotype None, but any child tier should have a tier type with any of the other options in the list of Stereotypes, defining the type of relation between parent and child annotations. The available Stereotype options are described with example in the manual.

Based on your description I think that LI_TYPE (and probably all/most of the other types for child tiers) can have the Stereotype Symbolic Association, defining a one-to-one relation between parent and child annotation.


Thank you so much, Han! This was a very helpful response. I now understand what the issue was: I was attempting to not assign LI_TYPE a stereotype, because it is possible for language ideologies to span multiple SPEECH annotations (multiple turns in a conversation). I now see that all child tiers must have a stereotype.

As far as I can tell, none of the dependent-tier stereotypes are capable of spanning multiple annotations in the parent tier. The only solutions I can think of are to 1) reverse the relationship, i.e., make LI_TYPE the parent of SPEECH; or 2) limit LI_TYPE annotations to one SPEECH annotation, and use them to manually extract any multi-turn segments that might occur. Does this sound accurate to you?

Yes, that’s correct; in ELAN it starts with the larger units on top-level tiers which can be subdivided on dependent tiers lower in the hierarchy (unlike some other tools where you can start with the smallest units). So, depending annotations spanning multiple annotations on the parent tier are not supported in ELAN. Solution 1), reversing the hierarchy, sounds a bit artificial; I guess speech comes first and attribution of a language ideology follows. Therefore approach 2) seems more obvious; each speech annotation is coded with a language ideology entry from the CV and detection of multi-turn segments can be performed later manually or maybe with a search query.