To begin:
- Only a limited amount of users are accepted simultaneously on the same corpus.
- The IPP engine may implement multiple or single modality (e.g. keyboard or pen/mouse signals); and the UI will work depending on such engine capabilities.
Once you have read this page, go and try IPP-Ann!
IPP-Ann overview
Choosing corpora
First of all, you must choose one of the available corpora by clicking on the parse button at index page. By clicking on use custom server? label, you can specify a custom IPP server while your session is active.
The demos are defined in XML files, each one having a dedicated parsing server. If an XML file is not valid, you cannot work on that demo.
Once you have selected a corpus, you can parse interactively each sentence line by line.
Parse your own sentence
You can type a sentence and parse it with the IPP system (currently supported languages are English and Spanish).
Please note that this option is for testing purposes, so you won't be able to export your result. For achieving that goal, please consider uploading a file.
Parse your own corpus
You can try your own data with IPP-Ann by uploading a file in any of these formats:
- Plain text
- XML
- HTML
To upload a document, just browse for your file and click the upload button. The system will try to find the most suitable IPP server for your chosen language.
You must upload only UTF-8 encoded files in english or spanish.
The sentences will be extracted from each paragraph of text in your file.
After using IPP-Ann to parse your sentences, you can download the parse trees them via the download link below the sentence list. Note that only sentences you have actually interacted with are included.
Interface layout description
The top block element is the navigation menu. You can go to this website's index and/or read the usage instructions as you are working on each demo.
The center block element is the list of sentences to parse.
Tree operations
Four operations can be performed within IPP-Ann. Two of them modify tree constituents:
- span modification
- draw a line from the constituent node to the word that corresponds to the span's right index
- label substitution
- select the text field and type the correct one.
And the other two affect the number of unary productions stemming from a constituent:
- unary production insertion
- draw a line from the constituent node to the floating ball that appears below itself
- unary production removal
- reset the span of the constituent parenting the unary production (just draw a line from the constituent node to the word that corresponds to the span's right index)
Please note that:
- Modifiying the span of a constituent invalidates its label and makes the system to recalculate a new one.
- Modifying the label of a constituent validates the whole constituent (span plus label).
User interaction protocol
- First, select a sentence for annotation by clicking on a text line. The system will propose an inital, full parse tree for the input sentence. If you were working on that sentence previously, the translation will be loaded from a log file.
-
Then, find the first incorrect constituent in a left-to-right depth-first order and ammend it.
This operation validates the prefix tree (highlighted in green),
and recalculates the rest of the tree.
- If its span is incorrect, modify it accordingly. The label of the constituent will be recalculated.
- If the label is incorrect, modify it accordingly. The span of the constituent is unmodified and validated along with the newly introduced label.
- If the constituent parents too many or too few unary productions, insert or remove them (see operations section above) and modify the label of the newly introduced nodes accordingly
- Repeat until the tree has been correctly annotated. To fully validate the current annotation, press CTRL + Enter inside the textbox of any label.
Notes
- POS tags can be promoted to syntactic tags by increasing its span (and vice versa).
- Constituents' left span is modified by correcting the span of his left neighbouring constituent.
- Constituents can be inserted/deleted at the end of any subtree by modifying the span of left neighbours accordingly.
Using IPP-Ann with an electronic pen
If the server supports multimodality, you can modify the labels of constituents with any kind of pointing device, such as a stylus or a PC tablet. Just place your stylus over an incorrect constituent and write down the correct label.
You can also use your computer mouse, but this option is discouraged. That is because the computer mouse, unfortunately, is not too good at drawing with precision.
The application will remember the last interaction mode between trees, that is, if you validate a tree in the keyboard modality, the next tree will load in keyboard mode. Vice versa for the e-pen modality.
Other Operations
- zooming
- You can either use CTRL + mouse wheel or right click over the tree to display view options. Also you can use the keyboard shortcut CTRL + UP/DOWN.
- panning
- Just CTRL + click over the tree to move it over the screen. Also you can use the keyboard arrows UP/DOWN/RIGHT/LEFT.
- deleting logs
- Right click outside the tree boundaries and select Delete this log to delete the current tree's log or Delete ALL logs to delete all logs in current corpus. Also you can use the link placed at the bottom of the UI to delete all logs from current corpus.