Table Of Contents

1. General design information

1.1. Lexicon population

The selection of lexical entries had a twofold aim: (i) enlarging the number of entries based on BCs; (ii) ensuring representativity of Templates with, at least, 10 units each one.

With regard to (i), the strategy adopted was the one advocated in the Simple Manual Encoding. Accordingly, as was done for the first deliverable, given a certain BC, taken from the list of commonly adopted starting BCs, a lexical entry was created for each one of the usual wordforms expressing that BC. Besides, from those entries with different wordforms for the same BC, other lexical entries were created, namely those that correspond to other possible senses that more typically are associated to those wordforms. In the meantime, a record of so called dummy lexical entries - in fact, entries awaiting to be entered and encoded in the lexicon - was being kept. That record contains wordforms that express the meaning of the BC but that are not in the lexicon of PAROLE. Synonymous wordforms of the encoded BC wordforms, but under different senses from the BC at stake, were also included.

As for (ii), the main concern wrt selection was the existence of a syntactic counterpart in PAROLE lexicon.

After these two procedures, corpus frequency criteria have also been applied to complete the population.

Coverage and completeness

The expected level of coverage was accomplished with reference to BCs, Templates and Domains, although the percentage of predicative units is far larger than the minimal 25%. This is due to the large amount of predicative nouns in Portuguese.

As for completeness, there are some Templates such as Agentive or Telic that do not have the minimal number of 10 units as a result of their abstract meaning. Words with such a meaning are not very frequent. Nevertheless, they are represented by some semantic units.

Whenever frequent word meanings were not contemplated in the PAROLE syntactical encoding, the relevant syntactic descriptions were added in the PAROLE Lexicon.

* Use of background resources

Nothing to report.

1.2. Current Lexicon Contents

All the "minimally required information" features were encoded, as well as optional synonymy wrt BCs.
 

Note: see in section 3-Statistics, the tables stating SemUs per Template type, SemUs per Domain and SemUs per Semantic Class.

1.3. Remaining work

Preparation of the final deliverable, due to April 28^th.Some synonymy relations are still being added. For technical reasons, the connection between synonyms was left behind (the information was kept in a word file). The encoding procedure of synonymy in the Access tool becomes more practice after all words encoding.

LIS2 considered it important to check Template, Semantic Class and Domain assignment, ordering Usems by Template, by Semantic Class and by Domain, in order to assure coherence of criteria. This task will be finished until the end of April, as well as most predicative information checking.

2. Semantic encoding

2.1. Criteria for Syntax-Semantic linking

Syntax-Semantic linking was straightforward. The Reduced option was not necessary because the Determinant is a default position in Portuguese, so it was not considered as a syntactical position in Parole.

2.2. Criteria for assigning Domain Features

Whenever no label could be found for a Usem, the domain General was assigned. More than one Domain (normaly two or three) were assigned when the Usem was considered to belong to more than one field. Some systematic assignments can be found between Templates and Domains as, for instance, Template Psychological Event and Domain Psychology, or between Template Illness and Domain Medicine. It was noticed that the Domain list should be completed. The lack of fields such as, for example, Orthopedy or Pneumology obliged the lexicographer to chose an upper node.

2.3. Criteria for assigning Semantic Class and Template Type

The strategy recommended in the Linguistic Specifications was followed, namely the Template tests, and, in most cases, the assignment of the Template recommended Semantic Class. However, in several cases a more specific one substituted the recommended Semantic Class. For instance, the lexicographers decided to assign Cardinal Point to cardinal points instead of the Template recommended Location.

As for the adjectives, the distinction between the attributive and predicative uses was made based on the semantic evaluation of the adjective-noun sense: whenever there was a restriction of the set of entities denoted by the noun, there was an attributive use, otherwise there was a predicative use.

2.3.1. Language specific typing

Nothing to report.

2.3.2. Template subtyping for language specific encoding

Nothing to report.

2.3.3. Criteria for encoding Semantic Relations

The Semantic Relations were encoded following the standard requirements of the Templates.

For synonymy (treated for most BCs words) the standard replacing test for deciding on the relation between two items was done, following the strategy used in the previous deliverable. As for the synonymy introduction into the tool, LIS2 decided to make the relation between two or more synonym Usems after they had all been encoded, for technical reasons related to the access tool.

Qualia information waaas only encoded in an experimental procedure, for about 20 SemUs.

2.3.4. Criteria for encoding Derivation Relations (optional).

Nothing to report.

2.4. Classes derived from encoding

Nothing to report.

2.5. Representation of Predicative Information.

The treatment of predicative information included some improvements in the PAROLE syntactical encoding. New SynUs were created and new descriptions were added, in order to cover the entire polissemy of the MUs.

As stated in the guidelines, the deverbal and deadjectival nouns have as their predicate the respective verb or adjective. Whenever there is a non-deverbal or non-deadjectival predicative noun, it stands as its own predicate in a Master Type of Link. In these cases, the Arg0 position is not fulfilled.

2.6. Problems encountered.

At this stage, the main problem relates to the encorporation of SGML semantic information with the PAROLE SGML final code. The Access tool does not allow individual morphologic and syntactic units information linking automatically with semantic information.

3. Statistics

Usems per Template type

Usems per Domain

Usems per Semantic Class