25.1.3 Lexical Divergences

Lexical divergences also cause huge difficulties in translation. We saw in Ch. 20, for example, that the English source language word bass could appear in Spanish as the fish lubina or the instrument bajo. Thus translation often requires solving the exact same problems as word sense disambiguation, and the two fields are closely linked.

In English the word bass is homonymous; the two senses of the word are not closely related semantically, and so it is natural that we would have to disambiguate in order to translate. Even in cases of polysemy, however, we often have to disambiguate if the target language doesn't have the exact same kind of polysemy. The English word know, for example, is polysemous; it can refer to knowing of a fact or proposition (I know that snow is white) or familiarity with a person or location (I know Jon Stewart). It turns out that translating these different senses requires using distinct French verbs, including the verbs connaître, and savoir. Savoir is generally used with sentential complements to indicate knowledge or mental representation of a fact or proposition, or verbal complements to indicate knowledge of how to do something (e.g., WordNet 3.0 senses #1, #2, #3). Connaître is generally used with NP complements to indicate familiarity or acquaintance with people, entities, or locations (e.g., WordNet 3.0 senses #4, #7). Similar distinctions occur in German, Chinese, and many other languages:

(25.7) English: I know he just bought a book.

(25.8) French: Je sais qu'il vient d'acheter un livre.

(25.9) English: I know John.

(25.10) French: Je connais Jean.

The savoir/connaître distinction corresponds to different groups of WordNet senses. Sometimes, however, a target language will make a distinction that is not even recognized in fine-grained dictionaries. German, for example, uses two distinct words for what in English would be called a wall: Wand for walls inside a building, and Mauer for walls outside a building. Similarly, where English uses the word brother for any male sibling, both Japanese and Chinese have distinct words for older brother and younger brother (Chinese gege and didi, respectively).

In addition to these distinctions, lexical divergences can be grammatical. For example, a word may translate best to a different part-of-speech in the target language. Many English sentences involving the verb like must be translated into German using the adverbial gern; thus she likes to sing maps to sie singt gerne (SHE SINGS LIKINGLY).

In translation, we can think of sense disambiguation as a kind of specification; we have to make a vague word like know or bass more specific in the target language. This kind of specification is also quite common with grammatical differences. Sometimes one language places more grammatical constraints on word choice than another. French and Spanish, for example, marks gender on adjectives, so an English translation into French requires specifying adjective gender. English distinguishes gender in pronouns where Mandarin does not; thus translating a third-person singular pronoun tā from Mandarin to English (he, she, or it) requires deciding who the original referent was. In Japanese, because there is no single word for is, the translator must choose between iru or aru, based on whether the subject is animate or not.

The way that languages differ in lexically dividing up conceptual space may be more complex than this one-to-many translation problem, leading to many-to-many mappings. For example Fig. 25.2 summarizes some of the complexities discussed by Hutchins and Somers (1992) in relating English leg, foot, and paw, to the French jambe, pied, patte, etc.

Further, one language may have a lexical gap, where no word or phrase, short of an explanatory footnote, can express the meaning of a word in the other language. For example, Japanese does not have a word for privacy, and English does not have a word for Japanese oyakoko or Chinese xiáo (we make do with the awkward phrase filial piety for both).

25.2 CLASSICAL MT & THE VAUQUOIS TRIANGLE

The next few sections introduce the classical pre-statistical architectures for machine translation. Real systems tend to involve combinations of elements from these three architectures; thus each is best thought of as a point in an algorithmic design space rather than as an actual algorithm.

In direct translation, we proceed word-by-word through the source language text, translating each word as we go. Direct translation uses a large bilingual dictionary, each of whose entries is a small program with the job of translating one word. In transfer approaches, we first parse the input text, and then apply rules to transform the source language parse structure into a target language parse structure. We then generate the target language sentence from the parse structure. In interlingua approaches, we analyze the source language text into some abstract meaning representation, called an interlingua. We then generate into the target language from this interlingual representation.

A common way to visualize these three approaches is with Vauquois triangle shown in Fig. 25.3. The triangle shows the increasing depth of analysis required (on both the analysis and generation end) as we move from the direct approach through transfer approaches, to interlingual approaches. In addition, it shows the decreasing

amount of transfer knowledge needed as we move up the triangle, from huge amounts of transfer at the direct level (almost all knowledge is transfer knowledge for each word) through transfer (transfer rules only for parse trees or thematic roles) through interlingua (no specific transfer knowledge).

In the next sections we'll see how these algorithms address some of the four translation examples shown in Fig. 25.4