Moses: Open Source Toolkit for Statistical Machine Translation
Proceedings of the 45th Annual Meeting of the ACL on Interactive Poster and Demonstration Sessions - ACL '07, 2007 · paper · Computational linguistics
Describes Moses, a free open-source toolkit for statistical machine translation. It adds support for linguistic factors, confusion network decoding and efficient data formats, includes training and tuning tools, and is reported to match leading systems in quality and speed.
Checked: PDF from aclanthology.org, fetched 5 Oct 2026. Page numbers are those of that PDF.
2 findings
Reference or label Figure 2 (page 2): the morphology factor for the second-person pronouns "vous" and "you" is printed as "1st", the same as for "je" and "i".
- p. 2 Figure 2, French row, second column
vous PRO vous 1st
- p. 2 Figure 2, English row, third column
you PRO you 1st
Only the authors can say what is right here.
Judge's note: On the page image and the enlarged crop, the French vector reads (vous, PRO, vous, 1st) and the English vector reads (you, PRO, you, 1st), identical in the last row to (je, PRO, je, 1st) and (i, PRO, i, 1st). The row holds person, tense, number and gender values (1st/present, masc, sing), so "1st" denotes grammatical person. The label on these second-person pronouns looks copied from the first column. The figure is illustrative, so the authors should confirm the intended label (2nd).
Second check: another model agreed
Report a problem with this finding
Reference or label Page 4, Section 5: the citation "(Federico and Bertoldi 2006)" has no matching entry in the reference list; the only Bertoldi/Federico entry is "Bertoldi, Nicola, and Marcello Federico. 'A New Decoder for Spoken Language Translation Based on Confusion Networks.' ... (ASRU), 2005".
- p. 4 Section 5, last paragraph
performance (Federico and Bertoldi 2006).
- p. 4 References
Bertoldi, Nicola, and Marcello Federico. "A New De-
- p. 4 Section 5, last paragraph
performance (Federico and Bertoldi 2006)
- p. 4 Section 5, last paragraph
(Federico and Bertoldi 2006)
Only the authors can say what is right here.
Judge's note: Read the full reference list on the page 4 image (Axelrod to Zens). There is no Federico-first or 2006 Federico/Bertoldi entry. The 2005 ASRU entry is cited separately as "(Bertoldi and Federico 2005)" in Section 4 for confusion network decoding, a different topic from language model quantization. The authors need to supply the missing reference.
Second check: another model agreed
Report a problem with this finding
11 minor (typos and the like; most checked by the judge only)
Wording Page 2: "Moses was the subject of this year's Johns Hopkins University Workshop on Machine Translation (Koehn et al. 2006)" — the paper is dated June 2007 and cites downloads "as of 1st March 2007", so the 2006 workshop is not "this year's".
- p. 2 Section 2, paragraph 5
year’s Johns Hopkins University Workshop on Machine Translation (Koehn et al. 2006)
- p. 2 Section 2, paragraph 4
1000 downloads as of 1st March 2007
- p. 4 References, Koehn et al.
Report of the 2006 Summer Workshop at Johns Hopkins University, 2006.
- p. 2 Section 2, paragraph 5
year’s Johns Hopkins University Workshop on
Judge's note: Page 2 image shows "this year's ... (Koehn et al. 2006)" directly after "1000 downloads as of 1st March 2007". The page 1 footer reads "Prague, June 2007". The page 4 reference reads "Report of the 2006 Summer Workshop at Johns Hopkins University, 2006". No text explains the wording, so it is a relative-date inconsistency, probably left over from a draft. The meaning is still recoverable.
Second check: another model agreed
Report a problem with this finding
Wording Figure 1 translates English "i am buying you a green cat" into French, while Figure 2 has the arrow pointing from the French factor vectors (je vous achet un chat) down to the English ones (i buy you a cat), with "green"/"vert" dropped; direction and example differ between the two figures.
- p. 2 Figure 1
i am buying you a green cat
- p. 2 Figure 2
i buy you a cat
- p. 2
⎟⎜ ⎟ ⎝ ⎠⎝ ⎠⎝ ⎠⎝ ⎠⎝ ⎠ 1 1 / 1 sing sing i buy you a cat PRO
Judge's note: Verified on the page 2 image. Figure 1 has English on top with arrows to French; Figure 2 has French on top with a downward arrow to English and a shorter sentence. The text does not say the two figures show the same example, so this could be a deliberate simplification; it is a minor consistency point only. The second reviewer note (repeated "sing") is an extraction artefact: in the image "sing" appears once in each of two separate vectors (a, cat).
Second check: another model agreed, code check
Report a problem with this finding
Wording "The language model on disk is also converted into this binary format" — no binary format has been introduced earlier in the paper, so "this" and "also" have no antecedent.
- p. 4 Section 5, language model paragraph
is also converted into this binary format
- p. 4 Section 5, paragraph 3 on the page
is also converted into this binary format, resulting
Judge's note: Viewed the crop of page 4 and searched the text layer: "binary" occurs only once in the whole article. The preceding sentence mentions only "a data structure for language models", and the phrase-table paragraph describes a prefix tree loaded on demand, never calling it a binary format. The meaning can be inferred (the on-disk form of the data structure), so this is a minor wording ambiguity rather than an error with an evident fix.
Second check: another model agreed
Report a problem with this finding
Typo Page 1, Section 1: "It features all the capabilities of the closed sourced Pharaoh decoder" — "closed sourced" should be "closed source".
- p. 1 Section 1, paragraph 2
pabilities of the closed sourced Pharaoh decoder
- p. 1 Section 1
closed sourced Pharaoh decoder
Judge's note: Viewed the page 1 image; the wording is printed as quoted. The paper uses "open-source" elsewhere. Purely cosmetic.
Second check: none (judge only)
Report a problem with this finding
Typo Page 2, Section 3: "minimizing the number dependencies for each step" — the word "of" is missing ("number of dependencies").
- p. 2 Section 3, paragraph after Figure 2
number dependencies for each step
Judge's note: Page 2 image shows "minimizing the / number dependencies for each step." with no "of" on either side of the line break. Not an extraction loss.
Second check: none (judge only)
Report a problem with this finding
Typo Figure 2 (page 2): the French surface-form factor is printed as "achet", whereas Figure 1 gives the same verb as "achète".
- p. 2 Figure 2, French row, third column
achet
- p. 2 Figure 1
achète
Judge's note: The crop shows the third French vector as (achet, VB, acheter, 1st / present). Figure 1 on the same page shows "achète" twice. The top row holds surface forms (je, vous, un, chat) and the lemma "acheter" has its own row, so "achet" is a truncated or misspelled surface form, not a deliberate stem.
Second check: none (judge only)
Report a problem with this finding
Typo Page 3, Section 3: "there is no decoding step which create source factors from other source factors" — singular subject with plural verb; should be "creates".
- p. 3 Section 3, paragraph 2 on page
which create source factors from other source fac-
Judge's note: Page 3 image shows the wording as quoted. Cosmetic grammar slip.
Second check: none (judge only)
Report a problem with this finding
Typo Page 3, Section 3: "it is possible to create a higher order language models for these factors" — singular article "a" with plural "models".
- p. 3 Section 3, paragraph 3 on page
it is possible to create a higher order lan-
- p. 3 Section 3
create a higher order lan-guage models
- p. 3 Section 3
create a higher order lan-
Judge's note: Page 3 image shows "create a higher order lan-/guage models for these factors". Cosmetic. The "which create" point in reviewer note 4 is handled under C010.
Second check: none (judge only)
Report a problem with this finding
Typo Page 4, References, Bilmes and Kirchhoff: venue printed as "HLT/NACCL, 2003", whereas the Koehn, Och and Marcu and the Zens and Ney entries print "HLT/NAACL".
- p. 4 References, Bilmes and Kirchhoff
HLT/NACCL, 2003.
- p. 4 References, Koehn, Och and Marcu
HLT/NAACL, 2003.
- p. 4 References, Bilmes and Kirchhoff
HLT/NACCL, 2003
- p. 4 References, Koehn, Och and Marcu
HLT/NAACL, 2003
Judge's note: Page 4 image shows "HLT/NACCL, 2003." in the Bilmes and Kirchhoff entry and "HLT/NAACL" in two other entries. The inconsistency is internal to the paper, so the transposition is evident without outside knowledge.
Second check: none (judge only)
Report a problem with this finding
Typo Page 4: "1.7 gigabytes to less than 20 mega bytes" — the unit is written as one word and then as two; should be "megabytes".
- p. 4 Section 5 continued, first paragraph
1.7 gigabytes to less than 20 mega bytes
Judge's note: Page 4 image shows a clear space in "mega bytes" beside "gigabytes" in the same sentence. Cosmetic.
Second check: none (judge only)
Report a problem with this finding
Typo Shen et al. reference prints the acronym as "The JHU Workshop 2006 Iwslt System." instead of "IWSLT".
- p. 4 References, Shen et al.
Iwslt System.
Judge's note: Viewed the crop of the page 4 reference list: the title reads "Iwslt System." while other acronyms in the list (JHU, BLEU, SRILM) are in capitals and the same entry spells out International Workshop on Spoken Language Translation. Cosmetic capitalisation slip.
Second check: none (judge only)