About analysis view Context Previous Next Query builder SVG tree Bracketed tree XML tree
The “default” view is a graphical presentation of the parse tree. If the sentence has been accessed through a query, the focus of the query is highlighted in the displayed tree.
By default the analysis view presents an analysis of one sentence. It is possible to extend the analysis presentation to include immediately preceeding or immediately following sentences by clicking respectively (when available) the leftside up arrow, or the leftside down arrow. If analyses for more than one sentence are being shown, then it is possible to remove the first sentence from view with a rightside down arrow, or remove the last sentence from view with a rightside up arrow. From a display of multiple sentences, clicking on a sentence id changes the view to that isolated example. Non-adjacent sentences can also be displayed, by clicking the “Context“ button to open a view that permits sentences to be selected freely from the wider context.
A tree view also gives navigation options:
At the bottom of the page, you can access alternative views of the annotation from a drop-down menu of choices starting with “default”. Alternatives choices are: “indexed”, “span”, and “formula”. After making a selection, click [Change mode].
A change to the “indexed” mode gives a view of the tree structure with indexing information that specifies argument relationships and antecedence relationships (including big PRO). This gives explicit indexing of grammatical dependencies that the annotation leaves implicit. Specifically: Indexing assigns a number in red above any node that serves as an argument/modifier for a predicate. The arguments/modifiers that a predicate takes are marked in brown under the terminal node for the predicate and mark the start of grey dotted lines that connect to the contributing argument/modifier. A zero element or pronoun that serves as the target for control or pronominalisation is indexed in blue below the node label of its immediately dominating phrase. A pink dotted line is drawn between the target and its antecedent.
Also, in the indexed mode, conjunction is expressed overtly (rather than implicitly) by modifying the structure of the tree. For clause conjunction, when an overt subject does not appear in a non-final conjunct, a VP layer is created for all conjuncts. If an overt subject appears in a non-final conjunct, an IML layer is created for all conjuncts.
The “span” mode shows relations of phrases to a target word. A sentence may contain several target words. Each target word is displayed on its own line. The relationship of a phrase to its target is specified by a label. By default, analyses for multiple sentences are displayed together. Clicking the [Sentence display] button displays results sentence by sentence.
The “formula” mode shows a predicate logic based formula as a semantic representation that is generated from the grammatical relations expressed in the tree. The source text is displayed above the formula. It is also possible to select [TPTP] for a rendering of the formula that could be sent to a theorem prover (Sutcliffe 2009). There are also further buttons that allow exploring the steps of transformation that are taken to reach the semantic representation, starting with the source annotation ([0]), followed by tree normalisation ([1]), conversion to a normalised Prolog term ([2]), conversion to an SCT expression ([3]), the initial result from semantic calculation ([4]), and the result from post compilation repositioning ([5]). See the Treebank Semantics documentation for details. Finally, there is a button to return to the initial formula view.
172 ted_talk_1
