Silei Xu
Silei Xu
in `test_postprocess_nlg_en.js`, the test may fail on a certain date. For example, if the test is run on Saturday, the 3rd example will output "It will be cloudy tomorrow in...
This includes automatically convert the SQL schema into a thingpedia manifest; extract parameter dataset from the database values. In general, the type is very limited in SQL, but this at...
Currently, the augmentation step will load all parameter values to memory, which could be an issue for some cases like QA over Wikidata. We could either use mmap or do...
Entity strings in the sentences are not quoted when generating mturk inputs.
Current POS tagger in genie is using [en-pos](https://www.npmjs.com/package/en-pos). It uses an inductive method, which is fast but not as accurate compared to sota neural approaches. Also, it only supports English....
In the current tokenization code, 6-digit number (e.g., 100000) is tokenized as `PHONE_NUMBER_x`. We should simply normalize the potential phone numbers and copy them over - let the parser decide...
"30minutes" is tokenized as "30m inutes"; "Search for comedy movies that are rated R." is tokenized as "search for comedy movies that are rated r." (no space between r and...
E.g., we should generate examples like "3 in the afternoon" mapped to `Time(15,0)`.
- subquery - count over compound types with filters