What assumption does an n-gram model make about word probabilities when n=1?

What assumption does an n-gram model make about word probabilities when n=1?

• A. The probability of a word depends on the entire previous text.
• B. The probability of a word depends only on the previous n words, with n set to 1.
• C. The probability of a word is independent of context.
• D. The probability of a word depends on its position in the document.

Answer: (B)

The main idea is that a unigram (n=1) language model uses no context from earlier words when predicting the next word. In an n-gram framework, the probability of the current word is conditioned on the previous n−1 words. When n is 1, there are zero previous words to condition on, so the next word is predicted based on its overall frequency in the language, not on surrounding text. The option about depending only on the previous n words with n set to 1 aligns with this setup: as you fix n=1, the model’s context is minimized to none, which means the word’s probability does not rely on preceding words. That captures the idea that context is eliminated in a unigram model, making it the best fit among the choices. The other statements don’t fit: relying on the entire previous text uses far more context than allowed; dependence on position in the document isn’t how these probabilities are modeled; and saying the probability is independent of context is essentially true for a unigram model, but the phrasing of linking it to the n-parameter formulation (n=1) is the intended tie-in in this question.

The main idea is that a unigram (n=1) language model uses no context from earlier words when predicting the next word. In an n-gram framework, the probability of the current word is conditioned on the previous n−1 words. When n is 1, there are zero previous words to condition on, so the next word is predicted based on its overall frequency in the language, not on surrounding text.

The option about depending only on the previous n words with n set to 1 aligns with this setup: as you fix n=1, the model’s context is minimized to none, which means the word’s probability does not rely on preceding words. That captures the idea that context is eliminated in a unigram model, making it the best fit among the choices.

The other statements don’t fit: relying on the entire previous text uses far more context than allowed; dependence on position in the document isn’t how these probabilities are modeled; and saying the probability is independent of context is essentially true for a unigram model, but the phrasing of linking it to the n-parameter formulation (n=1) is the intended tie-in in this question.