Next: Concluding Remarks Up: Illustration Previous: The Analysis of
Transfer examples
A unification grammar defines the transfer relation between logical forms of two languages. Like in generation, the `input attribute' is a logical form. Instead of strings, logical forms of the target language are generated.
For example the input
to transfer may be a feature structure such as figure 5 for a
sentence such as `The army opens fire at the civilians' where
contains the English logical form and 
the Spanish logical form
.
The bilingual grammar will apply its rules, testing after each application whether the value of the attribute subsumes the input feature structure. The value of the attribute will gradually be instantiated. At the end of the process, the system will test whether the input feature structure subsumes the value of the attribute . If this is the case, then the value of will be considered the output of transfer.
An example of a simple rule in PATR notation is given in Figure 6.
The integers 
are names of feature structures, where 
will
normally be used to represent the mother node and 
represent
the daughter nodes. Application of this rule to the feature
structure of Figure 4 results in the three instantiations in Figure 7.
Figure 7: Three instantiations
An example of the rule
for the first daughter will be a lexical entry and may look as in Figure 8.
The simple English expression `army' has to be translated as a complex
expression in Spanish: `fuerza militar'. The rule will look as in Figure 9 where it is
assumed that the construction is analysed in Spanish as an ordinary noun-adjective construction,
and where the logical form of the adjective takes the logical form of the noun as its argument.
The translation for `civilian' is defined in a similar rule (although the translation of
`number' is different).
Note that this example of complex transfer is similar to the famous `schimmel - white horse'
cases.
As a result of the rule applications the value of the attribute in Figure 5
will get instantiated to the feature structure in Figure 10,
from which the generator generates the
string `La fuerza militar rompio el fuego a la poblacion civil'.
For some transfer equivalences a rule such as in Figure 6 will be too
simplistic. The rule in Figure 11 is used to translate a logical form
such as 
into
the Dutch equivalent 
, as in for example `Minister Kok
likes the reforms' vs. `De hervormingen bevallen minister Kok'.
Note that the attribute 
contains the Dutch
logical form.
The logical forms that are encoded in MiMo2 are more complex than in the foregoing
examples. For example attributes for 
and 
may be present to represent
information about voice, tense and aspect. The example in Figure 11 shows that it is sometimes necessary
to alter the value of `voice'. The Dutch logical form 
with 
is
related to 
with 
. This rule may look as in Figure 12.
Figure 10: The target feature structure after transfer
This approach is not entirely without problems. It seems that some redundancies might be inevitable with the architecture proposed in section 2. A monolingual grammar defines a relation between strings and logical forms. It thus defines possible logical forms as well. Similarly, a bilingual unification grammar will have to define possible logical forms. The notion `possible logical form' will thus have to be defined in two (or more in case of a multilingual system) places. Note that this problem is a problem for all transfer systems; it is not restricted to our specific implementation of a transfer system (although the examples where the problem shows up may be specific to our grammars).
Some constructions such as control verbs and relative clauses may be represented using reentrancies; for example `the soldiers tried to shoot the president' may be represented by a feature structure where the first argument of `try' is reentrant with the first argument of `shoot', cf. Figure 13. The translation of such logical forms to Dutch equivalents can be defined as in rule 14. However, it is not clear that such reentrancies are always as local as in this case; if the reentrancies can be further away the transfer grammar will have to be complicated (eg. by a threading mechanism) to be able to translate such constructions, although a transfer writer would prefer the possibility to state that `reentrancies should be copied over' (cf. [52]).
Next: Concluding Remarks Up: Illustration Previous: The Analysis of
Gertjan van Noord
Thu Nov 24 19:09:23 MET 1994