Our tokenizer takes a raw text and splits tokens by their morphological aspects. It also groups tokens into sentences. Our tokenizer is based on the LDC tokenizer used for creating English Treebanks although it uses more robust heuristics. Here are some key features about our tokenizer.
- Emoticons are recognized as one unit (e.g.,
- Hyperlinks are recognized as one unit (
- Numbers consisting of punctuation are recognized as one unit (e.g.,
- Repeated punctuation are grouped together (e.g.,
- Abbreviations are recognized as one unit (e.g.,
- File extensions are not tokenized (e.g.,
- Units are tokenized (e.g.,
- Usernames including periods are recognized as one unit (e.g.,
TokenizerDemo shows how the tokenizer can be used in APIs.