English tectogrammatical layer in a nutshell



Node types

Types of edges

Node structure




Types of edges (relations between two nodes)

Each tree consists of nodes and edges. The edges themselves bear no description, but we can think of the tree as having several different types of edges according to the node type of the daughter node in the given relation.

The most common relation between two nodes is the dependency relation between a governing node and its modifier, e.g. the relation between yellow and shoe in a yellow shoe.

Some edges in the tree, however, are non-dependency edges. They occur in the following cases:

  • paratactic constructions
  • list structures
  • phrasemes and light verb constructions and some other complex predicates
  • rhematizers, disjuncts, all sorts of sentential particles
  • linguistic root of the sentence and the technical root of the sentence

Non-dependency edges

A non-dependency relation is between the root of a paratactic construction (all sorts of coordination and apposition) and the members of the paratactic structure as well as the modifiers of the paratactic structure. See Figure 1. All members of the paratactic structure have the attribute value is_member=1. Modifiers of paratactic structures (the ones that are perceived as modifying each member of the structure or the structure as a whole) are also governed by the paratactic root structure node, but lack the is_member attribute.

Also, the edge between the paratactic structure root node (nodetype=coap) and its mother node is a non-dependency edge. The same goes for the edge between the root of a list structure nodetype=list and its mother node and the edge between an atomic node (nodetype=atom) and its mother node, as well as for the nodes with the functors DPHR (phrasemes) and CPHR (nominal parts of light verb constructions).

The edge between the technical root node and the linguistic root of the sentence is also a non-dependency relation.

Figure 1. Besides Messrs. Cray and Barnum. The word Messrs. modifies the entire coordination. Please note that the preposition besides, being a function word, is not represented by its own t-node, but it is replicated inside each member of the coordination, interpreting the text as Besides Cray and besides Barnum.

Dual dependency in the free verb complement

Apart from the non-dependency relation, there is one case where the dependency is dual, and this is the free verb complement (functor=COMPL). Free verb complement is a verb complement that does not modify a copular verb, is not obligatory and modifies the governing verb as well as one of its arguments.

A node that represents the free verb complement always depends on a lexical verb. To indicate the semantic dependency of the free verb complement on an argument of this lexical verb, we draw a reference (a green arrow in the visualization, the compl.rf atributte) from the free verb complement node to that argument (Figure 2).

Figure 2

Examples of free verb complement:

  • -ing/-ed clause: TW Services Inc. posted a $3.3 million third-quarter net loss, compared with a $24.9 million profit, reflecting $60 million of expenses related to its much-publicized proposed takeover by Coniston Partners. Last month its president, John S. Donnelly, resigned citing "management differences" with Mr. Kurzweil. Rescue crews, however, gave up hope that others would be found alive under the collapsed roadway.
  • noun/adjective/participle: A part of the construction-oriented conglomerate SsangYong Group, it took over the dying Dong-A Motor Co. in 1986. In composite trading on the American Stock Exchange, Verit closed unchanged yesterday at $3.875 a share.
  • each: Banking stocks were the major gainers Monday amid hope that interest rates have peaked, as Deutsche Bank and Dresdner Bank added 4 marks each to 664 marks ($357) and 326 marks, respectively. Enviropact said the proceeds will be used as working capital for expansion and to pay its existing tax liability of about $1.5 million that was due Sept. 15.
  • as + noun/adjective/clause: Here are price trends on the world's major stock markets, as calculated by Morgan Stanley Capital International Perspective, Geneva. But he said IBM would have won the business anyway as a sale to a third party that would have then leased the equipment to the customer.
  • -self: Both estimates would mark declines from the 1988 net of $5.81 billion, or $9.80 a share, which itself was well below the record IBM set in 1984.

Ambiguous dependencies

The dependency relation of certain adjuncts (expressed by adverbs or prepositional phrases) is not always unambiguous: they do not necessarily modify only one element within the sentence but they can have a relation to several elements at the same time. Only one dependency relation can be represented by an edge. With adjuncts expressed by adverbs or prepositional phrases, it is often impossible to determine the appropriate dependency relation unambiguously. Only one (basic) dependency relation is represented, then, and no special attributes are established that would express other semantic relations. Like in the original Penn Treebank, subtrees whose dependency is perceived as ambiguous are placed on the position in the tree that is closer to the root (i.e. higher).