In a tri-gram model, which words are considered to predict the next word?

In a tri-gram model, which words are considered to predict the next word?

• A. The previous word.
• B. The next word.
• C. Look at the previous two words to predict the next.
• D. All words in the sentence.

Answer: (C)

Tri-gram models predict the next word using the two words that came just before it. This means you’re estimating p(next word | the two preceding words). That two-word context is why the option describing “look at the previous two words to predict the next” is the right choice. If you used only the previous word, you’d be in a bi-gram (two-word) setup; if you tried to base it on all words in the sentence, that’s not how tri-grams operate in practice. For example, after the sequence “I would,” the model looks at those two words to decide the most probable next word, such as “like” or “prefer,” based on training data.

Tri-gram models predict the next word using the two words that came just before it. This means you’re estimating p(next word | the two preceding words). That two-word context is why the option describing “look at the previous two words to predict the next” is the right choice. If you used only the previous word, you’d be in a bi-gram (two-word) setup; if you tried to base it on all words in the sentence, that’s not how tri-grams operate in practice. For example, after the sequence “I would,” the model looks at those two words to decide the most probable next word, such as “like” or “prefer,” based on training data.