Parts of Speech Tagging: Determining (i)–(iv) Capabilities

Parts of speech (POS) tagging is the NLP task of assigning a grammatical category (a POS tag) to each token in a text. In standard POS tagging, the tag for a word is chosen in context, meaning the sentence’s meaning and structure help disambiguate words that can take multiple POS categories. This context-dependent nature is why POS tagging is typically formulated as a sequence labeling problem (predict tags for all tokens, using neighboring tokens as evidence).

At a practical level, POS tagging supports:

• (i) choosing the POS “dynamically” for each word based on the sentence meaning/context,
• (ii) choosing the POS “dynamically” for each word based on the sentence structure/context,
• (iii) producing all possible POS tags for a given word (this is not the default POS-tagging output, but can be provided via lexicons/dictionaries or constrained candidate sets),
• (iv) “all of the above” only if (iii) is interpreted as “POS tagging pipeline can provide all candidate tags,” not merely the standard POS-tagging model output.

Mermaid overview of the core framing:

Below we formalize each option and connect it to common POS-tagging formulations.

Footnotes

1. Wikipedia — Parts of speech - https://en.wikipedia.org/wiki/Part_of_speech - Background on what POS tags represent. ↩

Key interpretation of POS tagging outputs

Standard POS tagging outputs one predicted tag per token for a provided sentence. The prediction is context-sensitive: models use surrounding tokens (and sometimes the whole sentence) to decide between ambiguous tags (e.g., words like can, like, run that can be noun/verb depending on context). This directly supports the “dynamic as per meaning” and “dynamic as per structure” interpretations.

We can treat “meaning” and “structure” as overlapping sources of context evidence:

• Sentence structure: syntactic cues from neighbors (e.g., determiners before nouns, auxiliaries before verb forms).
• Sentence meaning: broader contextual semantics often captured implicitly by the same context encoder that learns these syntactic/semantic correlations.

Both are “dynamic” because the predicted tag can change when the sentence changes. 2

Footnotes

1. Wikipedia — Part-of-speech tagging - https://en.wikipedia.org/wiki/Part-of-speech_tagging - Describes POS tagging as assigning tags to each word in context. ↩

2. Jurafsky & Martin (CS224N notes) — POS tagging & context sensitivity (widely used lecture source) - https://web.stanford.edu/~jurafsky/slp3/ - Context-dependent tagging overview. ↩

Option-by-option determination: (i)–(iv)

(i) parts of speech for each word dynamically as per meaning of the sentence

Answer: Yes (in standard POS tagging, context determines tags).
In sequence labeling POS tagging, each token’s POS tag is chosen conditioned on the input sentence context. When the sentence meaning changes (which typically also changes context), the most likely tag for an ambiguous word can change. This is precisely the motivation for context-dependent POS tagging. 2

Keyword links: POS tagging contextual disambiguation sequence labeling token

---

(ii) parts of speech for each word dynamically as per sentence structure

Answer: Yes (strongly).
POS tagging is commonly modeled as sequence labeling, where neighboring tags/tokens provide structural/syntactic cues that determine the correct POS category. For many languages and tagsets, structure-driven cues (e.g., determiners, auxiliaries, punctuation, word order patterns) are central evidence. 2

Keyword links: syntactic context tag sequence Markov assumptions CRF

---

(iii) all parts of speech for a specific word given as input

Answer: Not by default in standard POS tagging; it’s optional/adjunct.
Classic POS tagging tasks are defined as: given a sentence, predict the best tag per token, not “return all possible tags for a word.” However, you can support “all possible tags” by:

1. Lexicon/dictionary lookup (list all POS tags a word can take), or
2. Candidate generation (e.g., allowed tags from morphology or a tag dictionary), or
3. N-best / probability thresholding from a model (approximation of “all likely tags”).

But that behavior is not the standard definition/output of POS tagging itself; it is an added capability depending on system design. 2

Keyword links: tag dictionary candidate set N-best decoding morphology

---

(iv) All of the above

Best classification: Usually not strictly, unless (iii) is interpreted as “available via system extensions.”
If you interpret (iii) literally (“POS tagging determines all POS tags for a word”), then standard POS tagging does not guarantee that—it typically selects one best tag per token in a specific sentence.

Therefore:

• If the question is about what POS tagging determines in the standard task setting: (i) and (ii) yes; (iii) no; (iv) no.
• If the course/system question allows “POS tagging pipeline” to include lexicon-based candidate enumeration or N-best outputs: then you could argue (iv) yes as an engineered superset of POS tagging.

For typical academic NLP-task definitions, the correct choice is:

(iv) is not strictly correct; POS tagging natively supports (i) and (ii), while (iii) requires extra mechanisms. 2

Footnotes

1. Wikipedia — Part-of-speech tagging - https://en.wikipedia.org/wiki/Part-of-speech_tagging - Describes POS tagging as assigning tags to each word in context. ↩ ↩² ↩³ ↩⁴

2. Jurafsky & Martin (CS224N notes) — POS tagging & context sensitivity (widely used lecture source) - https://web.stanford.edu/~jurafsky/slp3/ - Context-dependent tagging overview. ↩

3. Stanford CS224N (or similar) — Sequence labeling framing - https://web.stanford.edu/~jurafsky/slp3/ - Covers labeling each token in a sequence. ↩

4. NLP resources on tagging vs candidate generation (tag dictionaries / morphology) - https://en.wikipedia.org/wiki/Morphology_(linguistics) - Morphological information is commonly used to determine possible POS tags/candidates. ↩ ↩²