{ "paper_id": "W98-0121", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T06:03:18.487283Z" }, "title": "A 'Tree Adjoining' Grammar without Adjoining The case of scrambling in German", "authors": [ { "first": "Gerard", "middle": [], "last": "Kempen", "suffix": "", "affiliation": { "laboratory": "", "institution": "Leiden University University of Koblenz-Landau", "location": { "postBox": "PO Box 9555 Rheinau l", "postCode": "NL-2300 RB, D-56075", "settlement": "Leiden, Koblenz", "country": "The Netherlands Germany" } }, "email": "kempen@rulfsw.fsw.leidenuniv.nl" }, { "first": "Karin", "middle": [], "last": "Harbusch", "suffix": "", "affiliation": { "laboratory": "", "institution": "Leiden University University of Koblenz-Landau", "location": { "postBox": "PO Box 9555 Rheinau l", "postCode": "NL-2300 RB, D-56075", "settlement": "Leiden, Koblenz", "country": "The Netherlands Germany" } }, "email": "" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "", "pdf_parse": { "paper_id": "W98-0121", "_pdf_hash": "", "abstract": [], "body_text": [ { "text": "The psycholinguistically motivated grammar formalism of Performance Grammar (PG, [Kempen 97] ) is similar to recent versions of Tree Adjoining Grammar (TAG; cf. [Joshi et al. 91] ) in several important respects. lt uses lexicalized initial trees; it generates derived trees synchronously linked to conceptual structures described in the same formalism (as in Synchronous T AGs [Shieber, Schabes 90]); and it factors dominance relationships and linear precedence in surface structure trees ( [Joshi 87]) .", "cite_spans": [ { "start": 76, "end": 80, "text": "(PG,", "ref_id": null }, { "start": 81, "end": 92, "text": "[Kempen 97]", "ref_id": null }, { "start": 161, "end": 178, "text": "[Joshi et al. 91]", "ref_id": null }, { "start": 491, "end": 502, "text": "[Joshi 87])", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "PG differs from recent TAG versions in that the adjoining operation and auxiliary trees are absent. Adjunction is replaced by a combination of substitution-the only composition operation-and a special linearization component that takes care of ordering the branches of derived trees in a global manner without re-arranging the derived structures. PG has been worked out for substantial fragments of Dutch, including the well-known cross-serial dependencies in self-embedded clauses. Here we will outline how PG deals with scrambling phenomena in German without invoking adjunction. For TAG treatments of these phenomena we refer to [Becker et al. 91 PG's lexicalized initial trees, called lexical frames, are 3-tiered mobiles. The top layer of a frame consists of a single phrasal node (called the 'root'; e.g. S, NP, ADJP, PP), which is connected to one or more functional nodes in the second layer (e.g., SUBJect, HeaD, Direct OB-Ject, CoMPlement, MODifier). At most one exemplar of a functional node is allowed in the same frame, except for MOD nodes, which may occur several times (indicated by the Kleene star: MOD*). Every functional node dominates exactly one phrasal node in the third ('foot') layer, except for HD which immediately dominates a lexical (part-of-speech) node. Each lexical frame is 'anchored' to a lexical item-a 'lemma' printed below the lexical 80 node serving as the frame's HeaD (Fig. l ) . Associated with nodes in the top and bottom layers are f eature matrices (not discussed here), which can be unified with other matrices as part of the substitution process. Unification always involves one root and one foot node of two different lexical frames (see the filled circles in Fig. 1 ). Only non-recursive unification is used.", "cite_spans": [ { "start": 632, "end": 649, "text": "[Becker et al. 91", "ref_id": null } ], "ref_spans": [ { "start": 1407, "end": 1416, "text": "(Fig. l )", "ref_id": null }, { "start": 1706, "end": 1712, "text": "Fig. 1", "ref_id": null } ], "eq_spans": [], "section": "", "sec_num": null }, { "text": "s ~ CMPR SUB) HD DOBJ MOO- jp t jp ADVJIPPIS J repa~ieren ~ NP NP ~0- D~O- PJO ADJP~P/S J. Jp L ADJP~P/S ni~nd ,_,1 Fahirad ~ 1 Hf' ART 1 das Fig. /. Simp/", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Left-to-right order of the branches of a lexical frame is determined by the 'linearizer' associated with a lexical frame. We assume that every lexical frame has a one-dimensional array specifying a fixed number of positions for foot nodes. For instance, verb frames (i.e., frames anchored to a verb) have an array whose positions can be occupied by a Subject NP, a Direct Object NP, the Head verb, etc. Fig. 2 shows 13 out of 14 slots where foot nodes of German verb frames can go. Tue positions numbered MI through MI 1 belono to the Midfield (Ger. Mittelfeld); BI and B2 make up the Backfield (Nachfeld) . Not shown is the single Forefield (Vorfelcf) slot Fl, located to the left of M 1. The annotations at the arcs denote possible fillers of the slots. For example, in a main clause the Head verb is assigned the first Midfield slot (MI); in a subordinate clause it goes to the last Midfield position (M 11 ). Subject NPs that could not enter the Forefield (e.g. in subordinate clauses) are placed in M2 if its head is a personal pronoun, in M3 otherwise. (Note that frames anchored to other parts of speech than verbs (NP, PP) have their own specialized linearization array.) The fillers listed in slots M2 through M7 represent the unmarked order of verbal arguments (cf. [Uszkoreit 87] ). They may be accompanied by additional constituents, in particular by modifiers and by arguments that, because of being in emphatic or contrastive focus, have been moved to the left (e.g. in weil er ein Fahrrad den Kindern versprichr, because-he-a-bike-the-children-promises). These companions are positioned after the 'standard' fillers (if any).", "cite_spans": [ { "start": 1276, "end": 1290, "text": "[Uszkoreit 87]", "ref_id": null } ], "ref_spans": [ { "start": 403, "end": 409, "text": "Fig. 2", "ref_id": "FIGREF2" } ], "eq_spans": [], "section": "", "sec_num": null }, { "text": "A key property of Jinearization in PG is that certain constituents may move out of their 'own' array and receive a position in an array located at a higher level. This is because, due to subcategorization features, a linearization array may be instantiated incompletely. For instance, if a verb takes a non-finite complement clause, then slots MI through M3 are missing from the cornplement's array. If. in addition, the complement is subjected to 'clause union', slots M4 through M7 are absent as weil. In such cases, verb argurnents and adjuncts that need to be expressed overtly, look for a slot higher up in the hierarchy of verb frames and get hold of the first (i.e. lowest) slot that is within scope. E.g., in da\u00df sie den Lehrer das Fahrrad nicht reparieren sah (that she didn't see the teacher repair the bike), den Lehrer and das Fahrrad occupy the same M7 slot, in order of increasing depth (Fig. 3) . Tue mechanism that controls the distribution of constituents over the slots of a linearization array. is modeled as a Finite-State Automaton (FSA). The FSA associated with a lexical frame traverses its array from left to right. At each slot, it inspects the set of constituents that are waiting for placement in the array, andinserts there any constituents meeting the place-ment conditions on that slot (see the labels on the edges of Fig. 2) .", "cite_spans": [], "ref_spans": [ { "start": 901, "end": 909, "text": "(Fig. 3)", "ref_id": null }, { "start": 1348, "end": 1355, "text": "Fig. 2)", "ref_id": "FIGREF2" } ], "eq_spans": [], "section": "", "sec_num": null }, { "text": "PG is capable of generating the mildly context-sensitive language anbncn. Fig. 4b illustrates a possible derivation of a 3 b 3 c 3 based on the lexical frames in Fig. 4a . Tue linearization array associated with ABC frames contains four slots SI \".54 to be filled, respectively, by constituents of type AP (any number, in arbitrary order), B, ABC, and CP. Furthermore, a subcategorization feature in the ABC foot node of the recursive ABC frame causes deletion of slot S 1 of the embedded ABC linearization arrays.", "cite_spans": [], "ref_spans": [ { "start": 74, "end": 81, "text": "Fig. 4b", "ref_id": null }, { "start": 162, "end": 169, "text": "Fig. 4a", "ref_id": null } ], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Certain scrambling phenomena in German are interpretable as a consequence of PG's linearization scheme. Consider sentence ( 1 ), from [Rambow 94), with two non-finite clauses embedded in one another:", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "[S[S das Fahrrad zu reparieren] zu versuchen] Rambow presents acceptability ratings for 30 scrambled versions of this sentence, viz. for all permutations in which the NPs precede the verbs they belong to. (Only five constituents are permutable: two NPs and three verbs.) See Table l for a selection from these data. Tue verbs versprechen and versuchen can take several types of complement in addition to the one exemplified in (1 ). Tue non-finite complement clause may be extraposed, i.e. put behind the finite verb in subordinate clauses (as in (2)). Moreover, it allows the so-called \"Third Construction\" where only part of the non-finite complement clause, including the infinitival verb, is extraposed. In the PG treatment of these constructions (iliustrated in Fig. 5 ), the linearization arrays play a crucial role. We assume that, in sentence (2), reparieren's linearization array has been instantiated from slot M4 onward, and in sentence (3) only from slot M8 onward. Moreover, versuchen's array has been truncated as weil and only contains slots MS through B2. This implies that, in (2), the direct object das Fahrrad could find a place in reparieren's array, whereas it was moved upward into the finite clause in (3). As stated above, it is a subcategorization feature of a complement-taking verb that controls how the complement's linearization array will be instantiated. Emphatic or contrastive focus is another factor causing a constituent to move upward. A focused constituent is assigned to early positions in a clause, e.g. M3 or M4. lf that position is not available at the clause level it belongs to, it moves into the array of a higher clause.", "cite_spans": [], "ref_spans": [ { "start": 275, "end": 282, "text": "Table l", "ref_id": null }, { "start": 767, "end": 773, "text": "Fig. 5", "ref_id": null } ], "eq_spans": [], "section": "", "sec_num": null }, { "text": "The position of the two infinitives with respect to one another rums out to be the major source of variation in acceptability. In all fully or marginally acceptable versions (\"ok\" or \"?\"): (A) the non-finite clauses are adjacent, or (B) they are discontinuous, with the complement-taking infinitive (zu versuchen) following its complement (zu reparieren). The structures depicted in Fig. 7 violate PG's linearization scheme because of an illegal attempt of zu reparieren to move into the fi-nite clause: this CMP-S is not moving into a focus slot and therefore will be assigned a place at its own level, i.e. in slot MS or B 1 of versuchen's array. All bad or unacceptable sentences in Table l suffer from this problem, while those rated good or marginal all adhere to PG's linearization scheme. Version 23, whose rating is relatively good although it manifests an illegal extraposition attempt, is the only exception.", "cite_spans": [], "ref_spans": [ { "start": 383, "end": 389, "text": "Fig. 7", "ref_id": null }, { "start": 686, "end": 693, "text": "Table l", "ref_id": null } ], "eq_spans": [], "section": "", "sec_num": null }, { "text": "We conclude that PG is capable of accounting for a considerable portion of the variance in the acceptability judgments reported by [Rambow 94] . This suggests that the combination of 'substitution + Jinearization FSA' in PG could serve as an alternative to 'adjunction + substitution' in TAG.", "cite_spans": [ { "start": 131, "end": 142, "text": "[Rambow 94]", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null } ], "back_matter": [], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "Long Distance Scrambling and Tree Adjoining Grammars", "authors": [ { "first": "T\"", "middle": [], "last": "Becker", "suffix": "" }, { "first": "A", "middle": [ "K" ], "last": "Joshi", "suffix": "" }, { "first": "", "middle": [], "last": "Rambow", "suffix": "" } ], "year": 1991, "venue": "Papers", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Becker, T\" Joshi, A.K., Rambow, 0. (1991). Long Distance Scram- bling and Tree Adjoining Grammars. In: Pa- pers presented to EACL9 l, Berlin.", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "Grammatical performance in human sentence production and comprehension. Ms, Leiden University", "authors": [ { "first": "A", "middle": [ "K" ], "last": "Joshi", "suffix": "" }, { "first": "A", "middle": [ "K" ], "last": "Joshi", "suffix": "" }, { "first": "K\"", "middle": [], "last": "Vijay-Shanker", "suffix": "" }, { "first": "D", "middle": [], "last": "Weir", "suffix": "" } ], "year": 1987, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Joshi, A.K. (1987). The relevance of Tree Adjoining Grammar to generation. In: Kempen, G. (Ed.), Natural language genera- tion. Dordrecht: Kluwer. [Joshi et al. 91] Joshi, A.K., Vijay-Shanker, K\" Weir, D. (1991). The convergence of mildly context-sensitive grammatical formalisms. In Sells, P., Shieber, S.M., Wasow, T. (eds.), The Mental Representation of Grammatical Relations. Cambridge MA: MIT Press. [Kempen 97] Kempen, G. (1997). Grammati- cal performance in human sentence produc- tion and comprehension. Ms, Leiden Univer- sity. [Rambow 94) Rambow, 0. (1994). Formal and computational aspects of natural language syntax. PhD Thesis, University of Pennsylva- nia.", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "Word Order and constituent structure in German", "authors": [ { "first": "S", "middle": [ "M" ], "last": "Shieber ; Sbieber", "suffix": "" }, { "first": "Y", "middle": [], "last": "Schabes", "suffix": "" } ], "year": 1987, "venue": "COL/NG-90", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Shieber, Schabes 90] Sbieber, S.M., Schabes, Y. (1990). Synchronous Tree-Adjoining Grammars. In: Karlgren, H. (Ed.), COL/NG- 90, Helsinki. [Uszkoreit 87] Uszkoreit, H. (1987). Word Or- der and constituent structure in German. Stanford CA: CSLI.", "links": null } }, "ref_entries": { "FIGREF0": { "text": "MainO:HD \"Q SUBJ/pers.pro ... 0 5ubordO: CMPR/c:onj V \u00ae SUBJ ... e OOBJ/pers. I refl.pro ... e TOBJ/pers. l refl.pro ... e IOBJ ... e OOBJ ... e non-finiteCMP-5 ...", "type_str": "figure", "uris": null, "num": null }, "FIGREF2": { "text": "Positions licensed to various rypes of constituents in the Midjield and Backfie/d of German clauses.", "type_str": "figure", "uris": null, "num": null }, "FIGREF3": { "text": "The embedded DOBJ-NP has been lifted into the linearization array (rectangle) of the next higher verb frame. Due to a subcategori;.ation feature of the lexica/ entry sehen (to see). only s/ors MB-MIO of the complement clause have been instantiated. This causes das Fahrrad to land in the M7 slot of the matrix. joining den Lehrer. Derivation of string a3 b3 c3. (a) Initial lexical frames. {b) Derived rree. Notice that only the marrix linearization array is instantiated completely; the embedded ones are trunc;ated, causing the A-phrases to be fronred.", "type_str": "figure", "uris": null, "num": null }, "FIGREF4": { "text": "PG analysis of sentence (3).", "type_str": "figure", "uris": null, "num": null }, "FIGREF5": { "text": "PG analyses of zwo acceptable utterances in conformiry with linearization rules. Top panel: both non-finite clauses occupy the Standardposition M8 in their respective arrays. NP das Fahrrad is focused (slot M3 or M4). Bottom panel: CMP-S versuchen is in unmarked position MB; CMP-S reparieren is focused.On the other band, in all unacceptable or bad versions (\"*\" or \"*?\"): (A') the non-finite clauses are discontinuous, (B') with the complement-taker preceding its complement. Examples are Rambow's sentences (10) and (30), quasi-reconstructed here asFig. 7.s Quasi-analyses of two unacceptable sentences.", "type_str": "figure", "uris": null, "num": null }, "TABREF2": { "text": "", "type_str": "table", "num": null, "html": null, "content": "
Acceprabiliry rarings for some scrambled ver-
sion of sentence ( 1 ), based on judgmenrs by several na-
tive speakers ofGennan. Data from (Rambow 941.
6 weil das Fahrrad zu reparieren niemandok
zu versuchen verspricht
20 weil niemand das Fahrrad zu reparieren?
verspricht zu versuchen
23 weil niemand zu versuchen verspricht,?
das Fahrrad zu reparieren
25 weil niemand das Fahrrad zu versuchen*?
verspricht zu reparieren
30 weil das Fahrrad zu versuchen niemand*?
verspricht zu reparieren
10 weil das Fahrrad zu versuchen niemand zu reparieren verspricht*
24 weil niemand zu versuchen das Fahrrad verspricht zu reparieren*
(1) weil niemand das Fahrrad zu reparieren
because nobody the bike to repair
zu versuchen verspricht
rorrypromises
'because nobody promises to try to repair
the bike'
(2) weil niemand verspricht das Fahrrad zu
reparieren zu versuchen
(3) weil niemand das Fahrrad verspricht zu
reparieren zu versuchen
" } } } }