Transliteration of text entry across scripts
Embodiments are disclosed for transliterating text entries across different script systems. A method according to some embodiments includes steps of: receiving an input string in a first script system input using a keyboard; segmenting, using a probabilistic model, the input string into phonemes that correspond to characters or sets of characters in a second script system; converting the phonemes in the first script system into the characters or sets of characters in the second script system, the characters or sets of characters forming a word or a word prefix in the second script system; and outputting the word or the word prefix in the second script system.
1. A computer-implemented method comprising:
receiving an input token written in a source script;
segmenting the input token into one or more segmentations of phonemes; for each segmentation:
determining a first probability that the segmentation is a valid segmentation of the input token;
determining a second probability of a phoneme in the source script being transliterated into a set of characters in a target script;
determining a third probability of the set of characters in the target script appearing in prior texts written in the target script; and
calculating a joint probability for the segmentation by combining the first, second and third probabilities;
selecting the segmentation having the highest joint probability; and
transliterating the token from the first script to the target script using the selected segmentation.
2. The method of claim 1 , wherein the joint probability, including the first, second and third probabilities, are calculated by a machine-learning model trained on a data set comprising a plurality of known transliteration conversions between the source script and the target script.
3. The method of claim 2 , the first probability being calculated by a segmentation likelihood module that determines a probability of a possible segmentation of the input token into a specific set of phonemes.
4. The method of claim 3 , the second probability being calculated by a decoding observation module that determines a probability of decoding the specific set of phonemes in the source script into a particular set of characters.
5. The method of claim 4 , the third probability being calculated by a prior distribution module that calculates a probability distribution over an output token.
6. The method of claim 1 wherein all possible segmentations of the input token are identified using a prefix tree model, wherein each path from a root node to a leaf node corresponds to a possible segmentation of the input token.
7. The method of claim 1 wherein all possible segmentations of the input token are identified using a directed acyclic graph, wherein each node in the graph represents a phoneme candidate at a specific location in the token.
8. A system comprising:
a processor; and
memory, coupled to the processor, the memory containing software that, when executed by the processor, causes the system to:
receive an input token written in a source script;
segment the input token into one or more segmentations of phonemes;
for each segmentation:
determine a first probability that the segmentation is a valid segmentation of the input token;
determine a second probability of a phoneme in the source script being transliterated into a set of characters in a target script;
determine a third probability of the set of characters in the target script appearing in prior texts written in the target script; and
calculate a joint probability for the segmentation by combining the first, second and third probabilities;
select the segmentation having the highest joint probability; and
transliterate the token from the first script to the target script using the selected segmentation.
9. The system of claim 8 , wherein the joint probability, including the first, second and third probabilities, are calculated by a machine-learning model trained on a data set comprising a plurality of known transliteration conversions between the source script and the target script.
10. The system of claim 9 , the first probability being calculated by a segmentation likelihood module that determines a probability of a possible segmentation of the input token into a specific set of phonemes.
11. The system of claim 10 , the second probability being calculated by a decoding observation module that determines a probability of decoding the specific set of phonemes in the source script into a particular set of characters in the target script.
12. The system of claim 11 , the third probability being calculated by a prior distribution module that calculates a probability distribution over an output token.
13. The system of claim 8 wherein all possible segmentations of the input token are identified using a prefix tree model, wherein each path from a root node to a leaf node corresponds to a possible segmentation of the input token.
14. The system of claim 8 wherein all possible segmentations of the input token are identified using a directed acyclic graph, wherein each node in the graph represents a phoneme candidate at a specific location in the token.
15. A non-transitory, computer readable medium containing instructions that, when executed by a processor, perform the functions of:
receiving an input token written in a source script;
segmenting the input token into one or more segmentations of phonemes;
for each segmentation:
determining a first probability that the segmentation is a valid segmentation of the input token;
determining a second probability of a phoneme in the source script being transliterated into a set of characters in a target script;
determining a third probability of the set of characters in the target script appearing in prior texts written in the target script; and
calculating a joint probability for the segmentation by combining the first, second and third probabilities;
selecting the segmentation having the highest joint probability; and
transliterating the token from the first script to the target script using the selected segmentation.
16. The medium of claim 15 , wherein the joint probability, including the first, second and third probabilities, are calculated by a machine-learning model trained on a data set comprising a plurality of known transliteration conversions between the source script and the target script.
17. The medium of claim 16 , the first probability being calculated by a segmentation likelihood module that determines a probability of a possible segmentation of the input token into a specific set of phonemes.
18. The medium of claim 17 , the second probability being calculated by a decoding observation module that determines a probability of decoding the specific set of phonemes in the source script into a particular set of characters in the target script.
19. The medium of claim 18 , the third probability being calculated by a prior distribution module that calculates a probability distribution over an output token.
20. The medium of claim 15 wherein all possible segmentations of the input token are identified using a prefix tree model, wherein each path from a root node to a leaf node corresponds to a possible segmentation of the input token.