What does a tri-gram model look at to predict the next word?

What does a tri-gram model look at to predict the next word?

• A. The previous two words.
• B. The previous one word.
• C. The next two words.
• D. All preceding words in the document.

Answer: (A)

A tri-gram model predicts the next word using the two words that come before it. The idea is that the word to come is influenced by a short window of context, specifically the two immediately preceding words, so it models P(next word | word1, word2). This is why the correct choice is that it looks at the previous two words. Looking at only the previous word would be a bigram model, and trying to predict two future words or using all preceding words would go beyond the fixed-context setup that defines n-gram models. In practice, this fixed two-word context captures common short-range patterns, though it can miss longer dependencies and can suffer from data sparsity without smoothing.

A tri-gram model predicts the next word using the two words that come before it. The idea is that the word to come is influenced by a short window of context, specifically the two immediately preceding words, so it models P(next word | word1, word2). This is why the correct choice is that it looks at the previous two words. Looking at only the previous word would be a bigram model, and trying to predict two future words or using all preceding words would go beyond the fixed-context setup that defines n-gram models. In practice, this fixed two-word context captures common short-range patterns, though it can miss longer dependencies and can suffer from data sparsity without smoothing.