Grapheme-to-Phoneme Conversion Quiz

Answers and explanations appear as you take the quiz.

1. In grapheme-to-phoneme conversion, what are 'phonemes'?
   • Distinct sound units of a language
   • Written letters of the alphabet
   • Whole words in a dictionary
   • Punctuation marks
2. Why is G2P especially hard for English?
   • English has no written form
   • English spelling is an inconsistent guide to pronunciation
   • English has only five phonemes
   • English words never repeat spellings
3. What is a 'heteronym' that G2P must resolve?
   • Two words spelled differently but sounding the same
   • A made-up brand name
   • A word with the same spelling but different pronunciations and meanings
   • A word borrowed from another language
4. Which resource is a classic English pronunciation dictionary used in G2P systems?
   • ImageNet
   • WordNet
   • GloVe
   • CMUdict
5. How do modern neural G2P models typically frame the task?
   • As sequence-to-sequence translation from letters to phonemes
   • As image classification
   • As clustering of audio clips
   • As a fixed one-letter-to-one-sound table
6. Why do many G2P systems still combine a dictionary lookup with a neural model?
   • Dictionaries are faster to train than anything else
   • The dictionary gives accurate pronunciations for common words while the model generalizes to unseen words
   • Neural models cannot output phonemes
   • Dictionaries store audio recordings directly
7. Why does a G2P model need a sequence alignment method such as attention or CTC?
   • Every written letter must map to exactly one phoneme
   • Phonemes are always longer than words
   • The input grapheme sequence and output phoneme sequence can have different lengths
   • Alignment is used only to identify speaker emotion
8. What does a stress marker such as ARPAbet AH0 versus AH1 encode?
   • The number of speakers in a recording
   • Whether a word is a proper noun
   • The language family of the spelling
   • A difference in the stress assigned to a vowel sound