Talbanken05

This is the home page for Talbanken05, a modernized version of Talbanken76, a Swedish treebank of roughly 300,000 words, constructed at Lund University in the 1970s. The treebank comes with no guarantee but is freely available for research and educational purposes as long as proper credit is given for the work done to produce the material (both in Lund and in Växjö). The archive available for download contains the entire treebank (divided into sections P, G, IB and SD) in three versions:
  1. MAMBA: Original syntactic and lexical annotation (original text encoding, with corrections)
  2. FPS: Flat phrase structure annotation (TIGER-XML encoding)
  3. DPS: Deepened phrase structure annotation (TIGER-XML encoding)
  4. Dep: Dependency structure annotation (Malt-XML encoding)
The two phrase structure versions are encoded in TIGER-XML, which means that they can be searched and displayed using TIGERSearch. The Dep version is encoded in Malt-XML but can be converted using MaltConverter to a TIGER-XML encoding of dependency structure conforming to the guidelines proposed by the Nordic Treebank Network.

Below we give a brief description of the original treebank (Talbanken76), the process of conversion, and the three different annotation standards (FPS, DPS, Dep). The conversion is described more fully in:

Talbanken76

Talbanken76 was originally published as: The data were collected in several projects at Lund University in the 1970s and the material is described in several publications: Teleman (1974) describes the analysis principles, while the other books apply these principles to different authentic materials.

Talbanken76 consists of a written language part and a spoken language part of roughly equal size. The written language part in turn consists of two sections, the so-called professional prose section (P), with data from textbooks, brochures, newspapers, etc., and a collection of high school students' essays (G). The spoken language part also has two sections, interviews (IB) and conversations and debates (SD). Altogether, the corpus contains close to 300,000 running tokens.

The MAMBA annotation scheme consists of two layers, the first being a lexical analysis, consisting of part-of-speech information including morphological features, and the second being a syntactic analysis, in terms of grammatical functions. Both layers are flat in the sense that they consist of tags assigned to individual word tokens, but the syntactic layer also gives information about constituent structure, as exemplified in the annotation of the sentence Genom skattereformen införs individuell beskattning av arbetsinkomster (Through the tax reform, individual taxation of work income is introduced): 

*GENOM                  PR        AAPR        
SKATTEREFORMEN          NNDDSS    AA          
INFÖRS                  VVPSSMPA  FV          
INDIVIDUELL             AJ        SSAT        
BESKATTNING             VN        SS          
AV                      PR        SSETPR      
ARBETSINKOMSTER         NN  SS    SSET        
.                       IP        IP

The first column of annotation is the lexical analysis, while the second column is the syntactic analysis. The grammatical subject of the sentence is the phrase individuell beskattning av arbetsinkomster (individual taxation of work income), where the head word beskattning (taxation) is assigned the simple tag SS for subject, while the pre-modifying adjective individuell (individual) is tagged SS and AT for adjectival modifier; in the post-modifying prepositional phrase, the noun arbetsinkomster (work income) is tagged SS and ET for post-modifier, while the preposition av (of) is tagged SS, ET and PR for preposition.

Tables explaining the categories used can be found here:

Conversion

The syntactic analysis in Talbanken76 is described by its creators as an eclectic combination of dependency grammar, topological field analysis and immediate constituent analysis. This makes it very suitable for conversion to both phrase structure and dependency annotation. The conversion has proceeded in three steps:
  1. The original flat but multi-layered annotation is converted to a bare phrase structure annotation, i.e. a phrase structure with unlabeled nonterminal nodes, and edges labeled with grammatical functions. This conversion is rather straightforward given the partially hierarchical annotation exemplified above.
  2. The bare phrase structure annotation is extended to a full phrase structure representation by labeling nonterminal nodes with syntactic categories. These categories are not part of the original annotation and have to be inferred from other parts of the annotation.
  3. The full phrase structure annotation is converted to a dependency annotation using the standard technique with head-finding rules and preserving grammatical functions as edge labels. Head-finding rules are not part of the original annotation scheme and have to be constructed manually.

Phrase Structure Annotation

The phrase structure annotation, which is the outcome of the second conversion step, uses a conventional set of phrase types (S, NP, VP, etc.) in combination with the grammatical functions of the original MAMBA annotation. The representation allows discontinuous phrases, although discontinuous constituents are relatively rare in the treebank.

The phrase structure annotation comes in two versions, one with the flattest possible trees that can be extracted from the original annotation, called Flat Phrase Structure (FPS), and one where trees have been deepened by inserting, e.g., NPs within PPs and VPs within (larger) VPs, called Deepened Phrase Structure (DPS). In both cases, the conversion has necessitated the introduction of a small number of new syntactic functions.

Dependency Structure Annotation

The dependency annotation (Dep), which is the outcome of the third conversion step, consists of terminal nodes connected by edges labeled with the same syntactic functions as DPS, extended with the label ROOT for words that are not governed by another word in the dependency structure. The representation allows non-projective dependency structures, which are needed to capture discontinuous constituents.

The conversion from phrase structure to dependency structure uses a priority list for finding the head of a phrase.