2.3 Polysemy and homonymy

Polysemy and Homonymy

Words frequently carry more than one meaning, and figuring out which meaning is intended is a core problem in semantics. Polysemy and homonymy are two ways this happens, and the distinction between them matters because it tells us something fundamental about how word meaning is organized. The key question is: are the multiple meanings related to each other, or not?

Polysemy vs. Homonymy

Polysemy occurs when a single word has multiple related meanings. The term comes from Greek: poly (many) + semy (meaning/sign). Because the meanings are related, they typically share a common etymological origin.

Take the word "head":

• The body part on top of your neck
• The leader of a group ("head of state")
• The top or front of something ("head of the line")

All three meanings connect back to a core idea of top or leading position. That shared thread is what makes this polysemy rather than coincidence.

Homonymy occurs when words share the same spelling and/or pronunciation but have unrelated meanings. Homonyms typically have different etymological origins; they just happen to look or sound alike. There are two subtypes:

• Homophones: same pronunciation, different spelling and meaning ("to," "too," "two")
• Homographs: same spelling, different pronunciation and meaning ("lead" as in to guide vs. "lead" the metal)

In practice, drawing the line between polysemy and homonymy isn't always clean-cut. Is "bank" (financial institution) vs. "bank" (river's edge) a case of homonymy or very distant polysemy? Most linguists treat it as homonymy because the meanings trace to different etymological roots, but borderline cases exist.

Identifying Lexical Ambiguity

Recognizing whether you're dealing with polysemy or homonymy comes down to testing whether the meanings are related.

Polysemy example: "Run" has a wide range of meanings, but they connect to a core sense of continuous movement or operation:

• Move quickly on foot
• Operate or function ("run a machine")
• Campaign for office ("run for president")
• Extend or spread ("colors running in the wash")

You can trace a thread from physical movement through all of these. That's polysemy.

Homonymy examples:

• "Bank" (financial institution) vs. "bank" (edge of a river): no meaningful connection between the two senses
• "Rose" (the flower) vs. "rose" (past tense of "rise"): completely different origins that happen to share a spelling

A useful test: if you can construct a plausible story for how one meaning extended from another, you're probably looking at polysemy. If the meanings feel like they belong to entirely different conceptual domains with no bridge between them, it's likely homonymy.

Context in Word Disambiguation

When you encounter an ambiguous word, you rely on two types of context to figure out the intended meaning:

• Linguistic context: the surrounding words, phrases, and sentences
• Situational context: the setting, participants, and purpose of the communication

Strategies for disambiguation:

1. Check the word's part of speech and syntactic role (this alone can resolve many cases)
2. Examine semantic relationships with nearby words (what topic is being discussed?)
3. Apply world knowledge to infer the most likely meaning

Compare these two sentences:

• "I need to go to the bank to deposit my paycheck." The words "deposit" and "paycheck" point clearly to the financial institution.
• "The fisherman sat on the bank of the river." Here, "fisherman" and "river" make the riverbank meaning obvious.

In most everyday conversation, disambiguation happens so fast you don't even notice it. The challenge arises when context is thin or when multiple meanings could plausibly fit.

Challenges for Language Processing

For computers, disambiguation is much harder than it is for humans. Word sense disambiguation (WSD) is the task of automatically determining which meaning of a word is intended in a given context. It matters for practical applications like machine translation and information retrieval, where picking the wrong sense leads to errors.

The main challenges include:

• Building comprehensive lexical resources (like WordNet) that capture the full range of a word's meanings
• Creating annotated training data where humans have labeled which sense is intended in thousands of examples
• Handling domain-specific meanings that shift depending on the field (e.g., "cell" in biology vs. telecommunications)
• Dealing with figurative language like metaphors and idioms, where literal word senses don't apply

Three broad approaches to WSD:

1. Knowledge-based methods: use structured resources like dictionaries and ontologies to match context to definitions
2. Supervised methods: train classifiers on human-annotated corpora where the correct sense is labeled
3. Unsupervised methods: cluster word occurrences by context patterns without pre-labeled data

For an intro course, the takeaway is this: what your brain does effortlessly with context remains a significant unsolved problem in computational linguistics.