If we use a reversible grammar for language understanding and generation then this implies that at least with respect to the grammatical knowledge these processes must be symmetric. This implication as been pointed out by [Appelt1987] and we can draw the organization of a reversible system as shown in figure 5.1.
In this graphic the conceptual component is divided into two parts: one used during understanding and one used during generation. This is done in order to emphasize the different tasks to be solved during both directions. It should not exclude the possibility that during both tasks same processes, knowledge sources or formalisms can be used. Furthermore we do not want to make any claims about the whole internal structure and status of the conceptual system so we view it as an open system (indicated by the dotted lines).
In this system we assume that semantic information is represented by some kind of logical formulas that are used to abstract predicate-argument structure and quantifier scoping from sentences. We need not to make further assumptions about the concrete form of the logical language in order to discuss the basic claims in this section. For utterances we make the simple assumption that they are represented as strings, i.e., we represent the phonological structure of a sentence as a list of words.
The graphic in figure 5.1 makes clear that the semantic representation of the grammar serves as an intermediate representation during understanding and generation. For the whole system we have a clean separation between conceptual and linguistic processing in both cases (indicated by the dashed line). The linguistic system has a modular status because it is the only component directly concerned with grammatical knowledge declaratively represented in a reversible grammar. This means that while performing both tasks the same grammatical power is potentially available (regardless of the actual language use). Clearly, such a modular design has the advantages already discussed in section 2.1.
Giving the linguistic component a modular status in that way, however, implies a serious problem especially for natural language generation, namely that the conceptual component constructs a logical form only on the basis of non-grammatical knowledge while the linguistic component processes logical forms only on the basis of grammatical knowledge. This means that the conceptual component cannot control completely whether and how the linguistic system will realize a given semantic structure.
For example, the following can happen. A message which is constructed precisely enough to satisfy the conceptual component's goal can be under-specified from the linguistic component's viewpoint. In particular, the generator can run into the risk of being misunderstood because of the produced utterance's ambiguity. We call this the choice problem of paraphrases.
For example, if the conceptual component specifies the following structure SEM
as input to the linguistic component:
then a possible utterance is `Remove the folder with the system tools' with the corresponding derived grammatical structure where the PP `with the system tools' is an adjunct to the VP:
From the generator point of view this utterance is grammatical and reflects exactly what the generator wants to express. For the hearer however there also exists the alternative grammatical structure where the PP `with the system tools' is a nominal adjunct:
with the semantic reading SEM':
The whole situation can graphically be represented as follows:
Remove the folder with the system tools
The left triangle represents the domain of the derivation between the semantic structure SEM and the utterance `Remove the folder with the system tools' obtained during generation. Both triangles represent the domain of derivation between the utterance and the semantic structures SEM and SEM' computed during parsing. Now the problem can be stated as follows:
Since during generation the linguistic component is mainly guided by the compositional structure of the semantic input, it cannot determine by itself those particular combinations of partial strings of the whole utterance which will lead to alternative derivations when the hearer is parsing this utterance. This means that possible ambiguities are out of the generator's view, and will only arise during parsing.
Of course, one could argue that if the generator had produced the utterance `Remove the folder by means of the system tools' instead of `Remove the folder with the system tools' then the kind of ambiguity exemplified above would not occur. Choosing the former instead of the latter in order to avoid ambiguity would mean that the conceptual component is able to foresee that the generation process will run into the risk of generating an ambiguity, and hence of conveying misinformation. The problem here depends on the alternative possible realizations of the instrument role, namely `with' or `by means of'. The conceptual component could have chosen `by means of' for some reasons internal to it (e.g., stylistic reasons, preferences, etc.) but not because it could foresee the ambiguity of `with'. In other words, given the modular design, the fact that at some point a potentially ambiguous LF surfaces as a non-ambiguous string cannot be assumed to be due to the fact that the ambiguity was foreseen, just other factors, independent from that, made the utterance unambiguous. If the conceptual component chooses `by means of' in order to restrict the set of possible derivations during parsing, this would mean that it is able to make decisions because of grammatical reasons.
The particular realization of the instrument role is not always relevant in order to avoid ambiguity. For example, in German (a language with relatively free word order) it would also be possible to utter:
`Mit den Systemwerkzeugen den Ordner löschen'
`[With the system tools] [the folder] remove'
(which can only mean Remove the folder by means of the system tools)
In this case the utterance is disambiguated by means of a specific ordering of the constituents. But now the same problem occurs: Without detailed grammatical background the conceptual component would not be able to specify the correct ordering in order to avoid ambiguity.
One might argue that when adding functional features to the feature system of a grammar (like focus, rhema, theme) in order to distinguish grammatical structures that have equal semantics but differ with respect to their functional value (cf. Fedder Fedder:91, Bateman et al. Bateman:92) the problem would not occur. Consider for instance the possible realizations of topicalization in German. Topicalization can be realized using either a passive construction or by fronting movement. Assume for instance that the conceptual component wants to verbalize the following semantic feature structure:
This semantic informationexpresses that `Peter is the one who drives Maria'.
The value of the FOCUS feature expresses that Maria
is the current focus of the communication situation. Possible
utterances in German are 1 and 2.
In both cases the syntactic construction can be marked by the two binary features FOCUS and EMPHASIS. In the passive case, the values of the FOCUS and EMPHASIS features would be defined as in the feature structure of (19) and for the fronting movement rule as represented in (20).
The problematic construction is that of (18) because during parsing there is also the unmarked reading possible, which says that Maria is the agent. Clearly, if the conceptual component wants to avoid misunderstandings it can choose the passive form. But to do this, it has to know that the value of EMPHASIS is necessary to distinguish both cases. But to have knowledge about this specific kind of combination of features means that it has to foresee that (18) is ambiguous. Hence it has to have detailed knowledge about the functional system and the way they are combined with specific constructions. Consequently it needs detailed knowledge about the actual grammar.
There is also psychologically grounded evidence for assuming that the
input to a tactical component might not be necessary and sufficient to
make linguistic decisions. This is best observed in examples of
self-correction [Levelt1989]. For example, in the following
utterance:
``but aaa, bands like aaa- aaa- aaa- errr- like groups, not bands, - groups, you know what I mean like aaa.''
the speaker discovers two words (the near-synonymous `group' and `band') each of which comes close to the underlying concept and has problems to decide which one is the most suitable. In this case, the problem is because of a mis-match between what the conceptual component want to express and what the language is capable of expressing [Rubinoff1988].
It is important to note here that the problem does not arise only when using reversible grammars but is an intrinsic problem of modularity. Every natural language model that assigns the grammatical component a modular status must face the problem. An important advantage of using reversible grammar is that we can consider the problem more clearly in the case of language processing. Moreover, in this thesis we demonstrate that a consistent use of reversible grammars is the starting point for solution of these problems.
Summarizing, it should be clear now that the conceptual component cannot have this kind of control because otherwise this would blur the modular design of a generation system mentioned above. Fortunately, in many situations of communication a speaker need not worry about the possible ambiguity of what she is saying because she can assume that the hearer will be able to disambiguate the utterance by means of contextual information or that she would otherwise ask for clarification (nevertheless, in the next chapter we show that the same problem mentioned above occurs also during clarification dialogs). However, an adequate generation system should also be able to avoid the generation of ambiguous utterances in some specific situations, e.g., when utterances refer to actions that have to be performed directly or in some specific dialog situations. As long as the conceptual component has no detailed knowledge of a specific grammar it could not express `choose this particular form to avoid ambiguity'. Therefore it can happen that the intended message will not be conveyed.
Currently, in generation systems where a modular design is advocated the problems are sometimes `solved' in such a way that the conceptual component has to provide all information needed by the linguistic component to make decisions about lexical and syntactic choices [McDonald1983], [McKeown1985], [Busemann1990], [Horacek1990], [McKeown et al. 1990], [Dale1990]. As a consequence, this implies that the input to the linguistic component is tailored to determine a good sentence, making the use of powerful grammatical processes redundant. In such approaches, linguistic components are only front-ends and the conceptual component needs detailed information about the language to use.
Hence, they are not separate modules because they both share the grammar. As pointed out in Fodor Fodor:83 one of the characteristic properties of a module is that it is computationally autonomous. But a relevant consideration of computationally autonomy is that modules do not share resources (in our case the grammar).
In order to be able to handle these problems, more flexible linguistic components are necessary that are able to handle, e.g., under-specified input. In [Hovy1987], [Finkler and Neumann1989, Neumann and Finkler1990] and [Reithinger1991] approaches are described how such more flexible components can be achieved. A major point of these systems is to assume a bidirectional flow of control between the conceptual and the linguistic component.
The problem with systems where a high degree of feedback between the conceptual and the linguistic component is necessary in order to perform the generation task is that one component could not perform its specific task without the help of the other. If the linguistic component has a modular status as assumed in this thesis then it is important for a component's mode of operation that it is minimally affected by the output of another component. Levelt Levelt:89 argues that ``it makes no sense to distinguish a processing component A whose mode of operation is continuously affected by feedback from another component, B. In that case, A is not specialist anymore, it won't come up with the right result without the `help' of B.'' ([Levelt1989], page 15). If this is the case one should better describe A and B as being one component.
