<?xml version="1.0" standalone="yes"?>
<Paper uid="P91-1021">
  <Title>TRANSLATION BY QUASI LOGICAL FORM TRANSFER</Title>
  <Section position="4" start_page="0" end_page="162" type="metho">
    <SectionTitle>
2 CLE REPRESENTATION
LEVELS
</SectionTitle>
    <Paragraph position="0"> In this section we explain how QLF fits into the overall architecture of the CLE and in section 3 we discuss the reasons for choosing it for interactive dialogue translation.</Paragraph>
    <Section position="1" start_page="161" end_page="161" type="sub_section">
      <SectionTitle>
2.1 CLE Processing Phases
</SectionTitle>
      <Paragraph position="0"> A coarse view of the CLE architecture is that it consists of a linguistic analysis phase followed by a contextual interpretation phase. The output of the first phase is a set of alternative QLF analyses of a sentence, while the output of the second is an RQLF (resolved QLF) representation of the interpretation of an utterance: Sentence --linguistic analysis--~ QL Fs Q X, Fs ---contextual interpretation---*&amp;quot; R Q L F. Deriving a fairly conventional Logical Form (LF) from the RQLF is then a simple formal mapping which removes the information in the RQLF that is not concerned with truth conditions.</Paragraph>
      <Paragraph position="1"> Linguistic analysis and contextual interpretation each consist of several subphases. For analysis these are: orthography, morphological analysis, syntactic analysis (parsing), and (compositional) semantic analysis. Apart from the first, these analysis subphases are based on the unification grammar paradigm, and they all use declarative bidirectional rules.</Paragraph>
      <Paragraph position="2"> When the CLE is being used as an interface to a computerized information system (e.g. a database system), its purpose is to derive an LF representation giving the truth conditions of an utterance input by a user. The LF language is based on first order predicate logic extended with generalized quantifiers and some other higher order constructs (Alshawi and van Eijck, 1989). For example, in a context where she can refer to Mary Smith, and one to &amp;quot;a car&amp;quot;, a possible LF for She hired one is: quant (exists ,C, \[carl ,C\], quant (exists ,E, \[event ,El, \[past, \[hir * I, E, mary_smith, C\] \] ) ).</Paragraph>
      <Paragraph position="3"> This can be paraphrased as &amp;quot;There is a car C, and an event E such that, in the past, ~. is a hiring event by Mary Smith of e.&amp;quot; In this notation, quantified formulae consist of a generalized quantifier, a variable, a restriction and a scope; square brackets are used for the application of predicates and operators to their arguments. To arrive at such LF representations, a number of intermediate levels of representation are produced by successive modular components.</Paragraph>
      <Paragraph position="4"> Generation of linguistic expressions in the CLE takes place from QLFs (or from RQLFs by mapping them to suitable QLFs). Since the rules used during the analysis phase are declarative and bidirectional, these are also used for generation.</Paragraph>
      <Paragraph position="5"> To achieve computationally efficient analysis and generation, the rules are pre-compiled in different ways for application in the two directions. Generation uses the semantic-head driven algorithm (Shieber et al, 1990).</Paragraph>
    </Section>
    <Section position="2" start_page="161" end_page="162" type="sub_section">
      <SectionTitle>
2.2 The QLF Language
</SectionTitle>
      <Paragraph position="0"> The QLF representations produced for a sentence are neutral with respect to the choice of referents for pronouns and definite descriptions, and relations implied by compound nouns and ellipsis. They are also neutral with respect to other ambiguities corresponding to alternative scopings of quantifiers and operators and to the collective/distributive and referential/attributive distinctions. The QLF is thus the level of representation encoding the results of compositional linguistic analysis independently of contextually sensitive aspects of understanding. These aspects are addressed by the contextual interpretation phase which has the following subphases: quantifier scoping (Moran 1988), reference resolution (Alshawi 1990), and plausibility judgement.</Paragraph>
      <Paragraph position="1"> The QLF language is a superset of the LF language containing additional expressions corresponding, for example, to unresolved anaphors.</Paragraph>
      <Paragraph position="2"> More specifically, there are two additional term constructs (anaphoric terms and quantified terms), and one additional formula construct</Paragraph>
      <Paragraph position="4"> a_form(Category, Pred Var , Restriction).</Paragraph>
      <Paragraph position="5"> These QLF constructs contain syntactic and morphological information in the Category and logical (truth-conditional) information in the Restriction, itself a QLF formula binding the variable. A QLF from which the LF for She hired one could have been derived is: \[past, \[hire, q_term (&lt;t =quant, n=s ing&gt;, E, \[event, E\] ), a_term(&lt;t =ref, p=pro, l=she, n=sing&gt;, Y, \[female, Y\] ), q_t erm (&lt;t =quant, n=sing&gt;, C, a_f orm(&lt;t =pred, l=one&gt;, P, \[P.C\]))\]\].</Paragraph>
      <Paragraph position="6">  in which categories are shown as lists of feature-value specifications (the feature shown are t for QLF expression type, n for number, p for phrase type, and 1 for lexical information). The differences between the QLF shown here and the LF shown earlier are that the quantified terms have been scoped, the anaphoric term for she has been resolved to Mary Smith, and the anaphoric NP restriction implicit in one has been resolved using the predicate car.</Paragraph>
      <Paragraph position="7"> The RQLF representation of an utterance includes all the information from the QLF, together with the resolutions of QLF constructs made during the contextual interpretation phase. For example, the referent of an a_term is unified with the a_term variable.</Paragraph>
      <Paragraph position="8"> Some constraints on plausibility can be applied at the QLF level before a full interpretation has been derived. This is because most of the predicate-argument structure of an utterance has been determined at that point, allowing, in particular, the application of sortal constraints expected by predicates of their arguments. Sortal constraints cut down on structural (e.g. attachment) ambiguity, and on word sense ambiguity, the latter being particularly important for the translation application in the context of large vocabularies. null</Paragraph>
    </Section>
  </Section>
  <Section position="5" start_page="162" end_page="163" type="metho">
    <SectionTitle>
3 REPRESENTATION LEVELS
FOR TRANSFER
</SectionTitle>
    <Paragraph position="0"> The representational structures on which transfer operates must contain information corresponding to several linguistic levels, including syntax and semantics. For transfer to be general, it must operate recursively on input representations. We call the level of representation on which this recursion operates the &amp;quot;organizing&amp;quot; level; semantic structure is the natural choice, since the basic requirement of translation is that it preserves meaning. null Syntactic phrase structure transfer, or deepsyntax transfer (e.g. Thurmair 1990, Nagao and Tsujii 1986) results in complex transfer rules, and the predicate-argument structure which is required for the application of sortal restrictions is not represented.</Paragraph>
    <Paragraph position="1"> McCord's (1988, 1989) organizing level appears to be/hat, of surface syntax, with additional deep syntactic and semantic content attached to nodes.</Paragraph>
    <Paragraph position="2"> As we have argued, this level is not optimal, which may be related to the fact that McCord's system is explicitly not symmetrical: different grammars are used for the analysis and synthesis of the same language, which are viewed as quite different tasks. Isabelle and Macklovitch (1986) argue against such asymmetry between analysis and synthesis on the grounds that, although it is tempting as a short-cut to building a structure sufficiently well-specified for synthesis to take place, asymmetry means that the transfer component must contain a lot of knowledge about the target language, with dire consequences for the modularity of the system and the reusability of different parts of it. In the BCI, however, the transfer rules contain only cross-linguistic knowledge, allowing the analysis and generation to make use of exactly the same data.</Paragraph>
    <Paragraph position="3"> Kaplan et al (1989) allow multiple levels of representation to take part in the transfer relation. However, Sadler et al (1990) point out that the particular approach to realizing this taken by Kaplan et al has problems of its own and does not cleanly separate monolingual from contrastive knowledge.</Paragraph>
    <Paragraph position="4"> The CLE processing subphases offer three semantic representations of different depth as candidates for an appropriate transfer level, namely QLF, RQLF and LF. At the LF level, sortal restrictions can be applied, but the form of noun phrase descriptions used and also information on topicalization is no longer present; the LF representation is too abstract for transfer. On the other hand, not all the information appearing in the RQLF about how QLF constructs have been resolved is necessary for translation. Resolved referents are not an adequate generator input for definite descriptions in the target language, since the view of the referent in the source is lost during translation. Another case is that translation from resolved ellipsis can result in unwieldy target sentences. In arguing for QLF-level transfer, we are asserting that predicate-argument relations of the type used in QLF are the appropriate organizing level for compositional transfer, while not denying the need for syntactic information to ensure that, for example, topichood or the given/new distinction is preserved.</Paragraph>
    <Paragraph position="5"> Finally, in contrast to systems such as Rosetta (Landsbergen, 1986) which depend on stating rule by rule correspondences between source and target grammars, we wish to make the monolingual descriptions as independent as possible from the task of translating between two languages. Apart from  its attractions from a theoretical point of view, this has practical advantages in allowing grammars to be reused for different language pairs and for applications other than translation.</Paragraph>
  </Section>
  <Section position="6" start_page="163" end_page="166" type="metho">
    <SectionTitle>
4 QLF TRANSFER
</SectionTitle>
    <Paragraph position="0"> QLF transfer involves taking a QLF analysis of a source sentence, say QLF1, and deriving from it another expression, QLF2, from which it is possible to generate a sentence in the target language. Leaving aside unresolved referential expressions, the main difference between QLF1 and QLF2 is that they will contain constants, particularly predicate constants, that originate in word sense entries from the lexicons of the respective languages.</Paragraph>
    <Paragraph position="1"> If more than one candidate source language QLF exists, the appropriate one is selected by presenting the user with choices of word sense paraphrases and of bracketings relating to differences in the syntactic analyses from which the QLFs were derived. null A transfer rule specifies a pair of QLF patterns.</Paragraph>
    <Paragraph position="2"> The left hand side matches QLF expressions for one language and the right hand side matches those for the other:</Paragraph>
    <Paragraph position="4"> If the operator is == then the rule is bidirectional.</Paragraph>
    <Paragraph position="5"> Otherwise, a single direction of applicability is indicated by use of one of the operators &gt;= or =&lt;.</Paragraph>
    <Paragraph position="6"> Transfer rules are applied recursively, this process following the recursive structure of the source QLF. In order to allow transfer between structurally different QLFs, rules with 'transfer variables' need to be used. These variables, which take the form tr(atom), show how subexpressions in the source QLF correspond to subexpressions translating them in the target QLF. For example, the following rule expresses an equivalence between the English to be called (&amp;quot;I am called John&amp;quot;), and the Swedish beta (&amp;quot;Jag heter John&amp;quot;).</Paragraph>
    <Paragraph position="8"> \[heCal, Cr (ev), tr (ag), Cr (name) \] ).</Paragraph>
    <Paragraph position="9"> Transfer rules often correspond directly to inter-lingual meaning postulates: when the expressions in a transfer rule are formulae, the symbols ==, &gt;=, and =&lt; can be read as the logical operators &lt;--&gt;, --&gt;, and &lt;-- respectively. A rule like Crans (\[and, \[bafll ,X\], \[luckl ,X\]\] \[otur I, x\] ) translating between the English bad luck and the Swedish otur, can be interpreted in this way.</Paragraph>
    <Paragraph position="10"> We will now assess the method's strengths and weaknesses, as they have manifested themselves in practice. We will pay particular attention to the criteria of expressiveness, compositionality, simplicity, reversibility and monotonicity.</Paragraph>
    <Paragraph position="11"> We take the last point first, since it is the most straightforward one. Since rules are applied purely nondeterministically and by pure unification, we get monotonicity &amp;quot;for free&amp;quot; - although there is a case for disallowing transfer by decomposition of a complex QLF structure which directly matches one side of a transfer rule. The other points need more discussion.</Paragraph>
    <Section position="1" start_page="163" end_page="164" type="sub_section">
      <SectionTitle>
4.1 Expressiveness
</SectionTitle>
      <Paragraph position="0"> Since we are intentionally limiting ourselves by not allowing access to full syntactic information (but only to that placed in QLF categories) in the transfer phase, it is legitimate to wonder whether the formalism can really be sufficiently expressive.</Paragraph>
      <Paragraph position="1"> Here, we will attempt to answer this criticism; we begin by noting that shortcomings in this area can be of several distinct kinds. Sometimes, a formalism can appear to make it necessary to write many rules, where one feels intuitively that one should be enough; we treat this kind of problem under the heading of compositionality. In other cases, the difficulty is rather that there does not appear to be any way of expressing the rule at all in terms of the given formalism. In our case, a fair proportion of problems that at first seem to fall into this category can be eliminated by having adequate mono-lingual grammars and using the target grammar as a filter; the idea is to allow the transfer component to produce unacceptable QLFs which are filtered out by fully constrained target grammars.</Paragraph>
      <Paragraph position="2"> A good example of the use of this technique is the English definite article, which in Swedish can be translated as a gender-dependent article, but preferably is omitted; however, an article is obligatory before an adjective. Solving this problem  (lit.: &amp;quot;has hurry&amp;quot;) at transfer level is not possible, since the transfer component has no way of knowing that a piece of logical form will be realized as an adjective; there are many cases where an adjective-noun combination in English is best translated as a compound noun in Swedish. Exploiting the fact that the relevant constraint is present in the Swedish grammar, however, the &amp;quot;transfer-and-filter&amp;quot; method reduces the problem to two simple lexical rules. Sortal restrictions at the target end can also be used as a filter in a similar way.</Paragraph>
    </Section>
    <Section position="2" start_page="164" end_page="164" type="sub_section">
      <SectionTitle>
4.2 Simplicity and reversibility
</SectionTitle>
      <Paragraph position="0"> The most obvious way to put the case with regard to simplicity is by giving a count of the various categories of rule, and providing evidence that there is a substantial proportion of rules which are simple in our framework, but would not necessarily be so in others.</Paragraph>
      <Paragraph position="1"> The transfer component currently contains 718 rules. 576 of these (80.2%) have the property that both the right- and left-hand sides are atomic.</Paragraph>
      <Paragraph position="2"> 502 members of this first group (69.9%) translate senses of single words to senses of single words; the remaining 74 (10.3%) translate atomic constants representing the senses of complex syntactic constructions, most commonly verbs taking particles, reflexives, or complementizers. An example is the following rule, which defines an equivalence between English care about ('John cares about Mary&amp;quot;) and Swedish bry sig om ( &amp;quot;John bryr sig om Mary&amp;quot;, lit. &amp;quot;John cares himself about Mary&amp;quot;).</Paragraph>
      <Paragraph position="4"> Since vocabulary has primarily been selected with regard to utility (we have, for example, made considerable use of frequency dictionaries (Alldn 1970)), we think it reasonable to claim that QLF-based transfer is simplifying the construction of transfer rules in a substantial proportion of the commonly encountered cases.</Paragraph>
      <Paragraph position="5"> On the score of reversibility, we will once again count cases; here we find that 659 (91.8%) of the rules are reversible, 17 (2.4%) work only in the English-Swedish direction, and 42 (5.8%) only in the Swedish-English direction. These also seem to be fairly good figures.</Paragraph>
    </Section>
    <Section position="3" start_page="164" end_page="166" type="sub_section">
      <SectionTitle>
4.3 Compositionality
</SectionTitle>
      <Paragraph position="0"> As in any rule-based system, &amp;quot;compositionality&amp;quot; corresponds to the extent to which it is necessary to provide special mechanisms to cover cases of irregular interactions between rules. As far as we know, there is no accepted benchmark for testing  compositionality of transfer; what we have done, as a first step in this direction, is to select six common types of complex transfer, and eleven common contexts in which they can occur. These are summarized in tables 1 and 2 respectively. Each complex transfer type is represented by a sample rule, as shown in table 1; the question is the extent to which the complex transfer rules continue to function in the different contexts (table 2).</Paragraph>
      <Paragraph position="1"> To test transfer compositionality properly, it is not sufficient simply to note which rule/context combinations are handled correctly; after all, it is always possible to create a completely ad hoc solution by simply adding one transfer rule for each combination. The problem must rather be posed in the following terms: if there is a single rule for each complex transfer type, and a number of rules for each context, how many extra rules must be added to cover special combinations? It is this issue we will address.</Paragraph>
      <Paragraph position="2"> The actual results of the tests were as follows.</Paragraph>
      <Paragraph position="3"> There were 124 meaningful combinations (some constructions could not be passivized); in 103 of these, transfer was perfectly compositional, and no extra rule was needed. For example, the English sentence for the combination &amp;quot;Verb to adjective + WH-question&amp;quot; is How much does John owe Mary.</Paragraph>
      <Paragraph position="4"> The corresponding Swedish sentence is Hut mycket dr John skyldig Mary? (&amp;quot;How much is John indebted-to Mary?&amp;quot;), and the two QLFs areS:  q_t erm(&lt;t =quanE, n=sing&gt;, A, \[state. A\] ), \[skyldiE_nsn_nst, a_t erm (&lt;t =ref, p=name&gt;, B, \[name_o~, B, j ohn3 ), a_term(&lt;t=ref, p=name&gt;, C, \[name_of, C,mary\] ), q_t erm(&lt;t =quant, l=wh&gt; ,D, \[quantity, D\] )\] \] \] \] It should be evident that the complex transfer rule defining the equivalence between owe and yarn skyldig, transC\[owe_have_to_pay, q_termC&lt;t=quant,n=sing&gt;,A,\[event,A\]), tr(ag),tr(sum),tr(obj)\] \[vara, q_term(&lt;t=quant,n=sing&gt;,A,\[state,A\]), \[skyldig_ngn_ngt, trCag),trCobj),tr(sum)\]\]).</Paragraph>
      <Paragraph position="5"> is quite unaffected by being used in the context of a Wit-question.</Paragraph>
      <Paragraph position="6"> Of the remaining 21 rule/context/direction triples, seven failed for basically uninteresting reasons: the combination &amp;quot;Perfect tense + Passiveto-active&amp;quot; did not generate in English, and the six sentences with the object-raising rule all failed in the Swedish-English direction due to the transfer component's current inability to create a function-application from a closed form. The final fourteen failures are significant from our point of view, and it is interesting to note that all of them resulted from mismatches in the scope of tense and negation operators.</Paragraph>
      <Paragraph position="7"> The question now becomes that of ascertaining the generality of the extra rules that need to be added to solve these fourteen unwanted interactions. Analysis showed that it was possible to add 26 extra rules (two of which were relevant here), which reordered the scopes of tense, negation and modifiers, and accounted for the scope differences between the English and Swedish QLFs arising from the general divergences in word-order and negation of main verbs. These solved ten of the outstanding cases. For example, the combination &amp;quot;Different particles + Negated&amp;quot; is John doesn't like Mary in English and John tycker inte om Mary (lit.: &amp;quot;John thinks not about Mary&amp;quot;) in Swedish; the QLF-pair is: \[pres p \[not, \[like, q_t erm(&lt;t=quant ,n=sing&gt;, A, \[event, A\] ), a_term ( &lt;t=ref, p=name&gt;, B, \[name_o~, B, j ohn\] ), a_termC&lt;t=ref, p=name&gt;, B, \[~ame_of, B ,mary\] )\] \] \]  \[not, \[present, \[tycka_om, q_t erm(&lt;t =quant, n=s ing&gt;, A, \[event, A\] ), a_t erm(&lt;t =ref, p=name&gt;, B, \[name_of, B, john\] ), a_term(&lt;t=ref, p=name&gt;, B, \[name_o:f, B, mary\] ) \] \] \] The extra rule here, trans( \[pres, \[not,tr(body)\]\] == \[not, \[present, tr (body)\] \] ).</Paragraph>
      <Paragraph position="8"> reorders the scopes of the negation and present-tense operators, but does not need to access the interior structure of the QLF (the &amp;quot;body&amp;quot; variable); this turns out to be the case for most interactions of negation, VP-modification and complex transfer. It is thus not surprising that a small number of similar rules covers most of the cases. The four bad interactions left all involved the English verb to be; these were the combinations &amp;quot;Passive to active / VP modifier&amp;quot; and &amp;quot;Idiomatic use of PP q- negation&amp;quot;, which failed to transfer in either direction. Here, there is no general solution involving the addition of a small number of extra rules, since the problem is caused by an occurrence of to be on the English side that is not matched by an occurrence of the corresponding Swedish word on the other. The solution must rather be to add an extra rule for each complex fransfer rule in the relevan~ class to cover the bad interaction. To solve the specific examples in the test set, two extra rules were thus required.</Paragraph>
      <Paragraph position="9"> Summarizing the picture, the tests revealed that all bad interactions between the transfer rules and contexts shown here could be removed by adding four extra rules to cover the 124 possible interactions. In a general perspective (viewing the rules as representatives of their respective classes), the rule-interaction problems exemplified by the concrete collisions were solved by adding * 26 general rules to cover certain standard scope mismatches caused by verb-inversion and negation.</Paragraph>
      <Paragraph position="10"> * two extra rules (one for present and one for past tense) for each complex transfer rule of either the &amp;quot;Idiomatic use of PP&amp;quot; or &amp;quot;Active to Passive&amp;quot; types, to cover idiosyncratic interactions of these with negation and VP-modification respectively.</Paragraph>
      <Paragraph position="11"> We view these results as very promising: there were few bad interactions, and those that existed were of a regular nature that could be counteracted without fear of further unwelcome sideeffects. This gives good grounds for hoping that the system could be scaled up to a practically useful size without suffering the usual fate of drowning in a sea of ad hoc fixes.</Paragraph>
    </Section>
  </Section>
  <Section position="7" start_page="166" end_page="166" type="metho">
    <SectionTitle>
5 IMPLEMENTATION STATUS
</SectionTitle>
    <Paragraph position="0"> The current implementation includes analysis, transfer, and generation modules, sizable grammars with morphological, syntactic and semantic rules for English and Swedish, and an experimental set of transfer rules for this language pair. Relative to the size of the grammars, the lexicons are still small (approximately 2000 and 1000 words respectively). About 250 entries for each language have been added for a specific domain (car hire), which makes possible moderately unconstrained conversation on this topic; the system, including the facilities for interactive resolution of translation problems, has been tested on a corpus of about 400 sentences relating to the domain. For short sentences typical of the car hire domain, median total processing times for analysis, transfer and generation are around ten seconds when running under Quintus Prolog on a SUN SPARCst~tion 2.</Paragraph>
    <Paragraph position="1"> We are currently investigating a different QLF representation of Iense, aspect and modality which should increase the transfer compositionality for the operator cases we have discussed in this paper, as well as allowing more flexible resolution of temporal relations in applications other than translation.</Paragraph>
  </Section>
class="xml-element"></Paper>