Google

An example of the 1-best selection process begins with a set of five candidate words, which are then ranked by probability: 'cute' (Pr = 0.34), 'on' (Pr = 0.32), 'sick' (Pr = 0.21), 'are' (Pr = 0.12), and '.' (Pr = 0.01). In the final output stage, the word 'cute' is selected because it has the highest score, while all other candidates are pruned.

Example of 1-Best Selection

A language model is generating the next word and has produced the following five candidate words with their associated probabilities: 'happy' (Pr = 0.41), 'sad' (Pr = 0.15), 'very' (Pr = 0.28), 'and' (Pr = 0.11), and 'is' (Pr = 0.05). Based on a process that selects only the single most likely candidate and discards the rest, identify which word is chosen for the output and which words are discarded.

Applying the 1-Best Selection Method

A language model is generating the next word for the phrase 'The dog is very...'. The model considers five potential words: 'friendly', 'big', 'running', 'brown', and 'asleep'. The final word chosen and output by the model is 'friendly'. If the model uses a selection method where only the single candidate with the highest probability score is chosen and all other candidates are discarded, what can you conclude about the probability score of 'friendly' in relation to the other four words?

Inferring Probability from Model Output

A language model is generating the next word in a sentence. After calculating the probabilities for several candidate words, it has the following set: 'the' (probability = 0.45), 'a' (probability = 0.25), 'my' (probability = 0.15), and 'your' (probability = 0.10). If the model uses a selection method where only the single most probable candidate is chosen and all others are discarded, which word will be selected for the final output?

Learn Before

Related