I wonder about the legitimacy of using the terms "POS tagging", "Chunking", "Disambiguation" and "Categorization" to describe an activity that doesn't include writing code and database queries, or interacting with the NLP algorithm and database directly.

More specifically, let's suppose I use the following tools:

  1. an "Annotator" for analyzing the input text (e.g. sentences copypasted from online newspapers) and choose and save proper values as regards to "POS" of tokens and words, "Words"(entities and collocations) and "Chunk". Tokens are already detected by default. I have to decide which words are entities and/or collocations or not and their typology, though. May the performed tasks be called "POS tagging", "Chunking" and "support to categorization"?

  2. A knowledge base, for searching and choosing the proper synsets of the lemmas and assigning them to the words analyzed in the previous Annotation tool. May such a task be called "Disambiguation"?

  3. A graphical user interface which shows how the NLP analyzes by default the input texts as regards to Lemmas, POS, Chunks, Senses, entities, domains, main concepts, dependency tree, in order to make analyses consistent with it.

If I want to define these activities in a few words, "Machine Learning annotation" may be the most correct.

But what if I want to be more specific? I don't know whether or not the terms "POS tagging", "Chunking", "Disambiguation" and "Support to categorization" may be appropriate for they generally come within "programming contexts", as far as I know. In other terms, do they involve writing algorithms and programming or are they / may they be referred to the "less-technical" activities described above?


The procedures you mention don't need to involve writing code. There are now many ready-made tools available which implement various algorithms.

POS-Tagging, Chunking, and Semantic Tagging/Disambiguation are knowledge-based procedures which can all be seen as classification/clustering tasks: POS-Tagging and Disambiguation are classifications, in that they propose a label (a POS- or semantic tag) that is assigned to a token. Chunking is a clustering algorithm, as it finds coherent groups in sequences of tokens. You could also view it as segmentation, as it splits a sentence into parts.

In principle they don't have anything to do with machine learning; the algorithm you are using (ML or not) is an implementation detail. Early Taggers were mostly rule-based, but with increasing availability of annotated data it became more feasible to use ML algorithms. This, however, has got nothing to do with how you classify the procedures.

In general I would call the activity you describe as 'annotation', as you enrich the source text by adding descriptive categories to tokens and token sequences.

So, no, they do not have to involve programming or implementing algorithms, and you don't need to be technical to perform them. Though some knowledge of linguistic concepts would help.

  • $\begingroup$ Than you for answering. What about the job role? May the person who performs these tasks ('annotation') be called "Computational Linguist"? $\endgroup$ – franz1 Feb 20 '20 at 15:22
  • 1
    $\begingroup$ Computational linguist is a much broader description, but yes, one aspect of working as a computational linguist could be to do annotation. $\endgroup$ – Oliver Mason Feb 20 '20 at 15:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.